effets de dopage, de réduction de taille et d'interface
TRANSCRIPT
HAL Id: tel-00952979https://tel.archives-ouvertes.fr/tel-00952979
Submitted on 28 Feb 2014
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.
Etude de matériaux pour mémoires à changement dephase : effets de dopage, de réduction de taille et
d’interfaceGiada Eléonora Ghezzi
To cite this version:Giada Eléonora Ghezzi. Etude de matériaux pour mémoires à changement de phase : effets de dopage,de réduction de taille et d’interface. Autre [cond-mat.other]. Université de Grenoble, 2013. Français.<NNT : 2013GRENY018>. <tel-00952979>
THESE
Pour obtenir le grade de
DOCTEUR DE L’UNIVERSITE DE GRENOBLESpecialite : Physique
Arrete ministeriel : 7 aout 2006
Presentee par
Giada Eleonora Ghezzi
These dirigee par Francoise Hippertet codirigee par Sylvain Maıtrejean
preparee au sein CEA Leti, Minatec campus et LMGP (CNRS, Grenoble-INP, Minatecet de Ecole Doctorale de Physique
Material studies for advancedphase change memories: doping,size reduction and interface effect
These soutenue publiquement le 25 fevrier 2013,devant le jury compose de :
Pr. Yves BrechetProfesseur Grenoble INP, President
Pr. Olivier ThomasProfesseur Universite d’Aix-Marseille, Rapporteur
Pr. David WrightProfessor University of Exter, Rapporteur
Dr. Christophe BicharaDirecteur de recherche CNRS, Examinateur
Dr. Paola ZulianiProject leader at ST Microelectronics, Examinatrice
Francoise HippertProfesseur Grenoble INP, Directeur de these
Sylvain MaıtrejeanIngenieur-chercheur CEA, Co-Directeur de these
To my beloved family
Contents
Summary 1
Resume 5
1 Phase Change Memories Overview 9
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2 Phase Change Memories working principle . . . . . . . . . . . . 12
1.2.1 Basic device example . . . . . . . . . . . . . . . . . . . . 15
1.2.2 Electrical conduction model . . . . . . . . . . . . . . . . 17
1.3 Physics of phase change transformations . . . . . . . . . . . . . 18
1.3.1 Amorphization . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.2 Crystallization . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.3 Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.4 Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.3.5 Johnson - Mehl - Avrami - Kolmogorov (JMAK) formalism 25
1.4 Phase change materials . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.1 Ge:Sb:Te compounds . . . . . . . . . . . . . . . . . . . . 27
1.4.2 Structure of crystalline and amorphous Ge2Sb2Te5 and GeTe 29
1.5 Goals and outline . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2 Effect of doping on the structure of amorphous GeTe 35
2.1 State of the art on doping effects in phase change materials . . . 36
2.2 Theory of the Pair Distribution Function (PDF) g(r) . . . . . . 40
2.3 Description of the samples . . . . . . . . . . . . . . . . . . . . . 45
2.4 X-Ray scattering measurements and results . . . . . . . . . . . . 46
2.5 Ab initio simulations . . . . . . . . . . . . . . . . . . . . . . . . 49
i
2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.7 Conclusions and perspectives . . . . . . . . . . . . . . . . . . . . 61
3 Confinement of phase change materials: Ge2Sb2Te5 nanoclusters 63
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.1.1 Effect of shrinking size in one dimension: thin films . . . 64
3.1.2 Effect of shrinking size in two and three dimensions: nanos-
tructures . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2 Clusters deposition . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3 X-Ray Diffraction study . . . . . . . . . . . . . . . . . . . . . . 81
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.5 Conclusions and perspectives . . . . . . . . . . . . . . . . . . . . 93
4 Interface effect on crystallization of PC thin films 95
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2 Reflectivity measurements . . . . . . . . . . . . . . . . . . . . . 97
4.3 X-Ray Diffraction measurements . . . . . . . . . . . . . . . . . . 103
4.4 Synchrotron X-Ray Diffraction . . . . . . . . . . . . . . . . . . . 110
4.5 Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . 112
Conclusion 121
A Experimental Techniques 125
A.1 Reflectivity measurements . . . . . . . . . . . . . . . . . . . . . 126
A.2 X-Ray Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . 127
A.2.1 Conventional X-Ray Diffraction laboratory experiment . 128
A.2.2 Large-scale facilities experiments . . . . . . . . . . . . . 129
B Deposition method 137
List of Figures 148
List of Tables 150
References 151
ii
Acknowledgments 165
iii
Summary
Phase Change Memories (PCM) are considered the best candidate for the
next generation of non volatile memories (NVM), which market is actually dom-
inated by Flash technology. PCM are based on the properties of some chalco-
genide materials, called phase change (PC) materials, to reversibly switch be-
tween a crystalline and an amorphous phase. These two phases are characterized
by very different electrical and optical properties, which makes possible to store
information. For PCM to be competitive a great research effort is still needed.
This effort should be directed to improve technological aspects such as device
architecture, layout and control circuitry, but also to optimize the PC materials
used in the memory cell. The effect of scaling and interface layers on the PC
materials properties must also be understood. Up to now the most used and
studied PC material is Ge2Sb2Te5 (GST), sometimes doped with nitrogen, but
research is extremely active in looking for other compounds that can offer better
properties. Materials to be used in PCM should crystallize fast, exhibit large
optical and electrical contrast between the amorphous and crystalline phases,
have a melting temperature sufficiently low to limit the electric power needed
for amorphization and their amorphous phase must be stable to grant good re-
tention performances. PC materials doping has been identified as a promising
solution for properties optimization. Moreover, those properties must not decay
after many transformation cycles between the amorphous and crystalline states
and must be kept with scaling, i.e. for a small amount of PC material embedded
1
in conducting and/or insulating materials. In this context, the aim of this work
is to investigate the effect of scaling and doping on PC materials properties.
Here, three directions have been explored. One is a contribution to understand
the impact of doping on the structure of amorphous GeTe. The second one is
an investigation of the effect of 3D confinement on the phase change in GST.
The third one is the study of influence of interface materials on crystallization
of GST and GeTe thin films.
In the first part of the thesis the local structural properties of C and N
doped amorphous GeTe are investigated through X-ray scattering experiments
performed at the synchrotron SOLEIL (Saclay, France). At the beginning of this
work it was demonstrated that C and N doping improves drastically the data
retention of GeTe and lowers the current needed for amorphization. The goal
was to understand the impact of doping on the amorphous structure of GeTe
by analyzing the pair distribution function of doped and undoped samples. The
impact of doping is revealed experimentally by the appearance of a new peak
in the pair distribution function of doped GeTe, indicating the formation of a
bond at a new distance, absent in the undoped amorphous material. Ab initio
simulations show the formation of new tetrahedral and triangular or pyramidal
environments centered on carbon or nitrogen as well as long carbon chains and a
few N2 molecules. The new peak observed experimentally corresponds to Ge-Ge
distances in the units centered on carbon or nitrogen. These structural changes
can be related to the enhanced crystallization temperature and activation energy
of C and N-doped GeTe.
The effect of confinement on the phase change in GST is the subject of the
second part of this work. The capability of PC materials to be scaled while main-
taining their properties is a fundamental requirement for further development of
PCM. There have been many studies on thin films of varying thickness. In some
cases the crystallization temperature has been reported to increase drastically
with reduced dimensions for films thinner than 10 nm, up to the point of losing
the phase change property for films thinner than 2 nm. These studies deal with
confinement in only one direction, the film thickness, but the ideal system for
studying the effect of scaling on PC materials in a memory cell (where both size
2
and interface effects play a role) is a set of nanoclusters. Indeed, the volume of
clusters is confined in three dimensions and the interface effects are enhanced
due to a larger surface/volume ratio. Nano-sized clusters of GST with an aver-
age size of around 5.7 nm (± 1 nm) in a matrix of Al2O3 have been made using a
sputtering gas phase condensation source and characterized by X-Ray diffraction
measurements performed at the ESRF synchrotron (Grenoble). At the moment,
those clusters are the smallest GST clusters ever deposited by sputtering, and
with the narrowest size distribution. These clusters have been made by a method
close to those used for PCM thin films deposition, thus giving information that
can be easily exported to device fabrication. The crystallization temperature
of clusters is around 180C, only slightly above the crystallization temperature
(155C ) of a 10 nm thin film of GST deposited under the same conditions and
embedded in Al2O3 . The crystalline phase is the cubic metastable phase of
GST for both clusters and thin film. The lattice parameter is larger in clusters
than in thin films. The lattice variation can be explained by supposing that
the surrounding rigid Al2O3 matrix exerts a tensile strain on the clusters and
that their volume during crystallization is forced to remain constant and equal
to their volume in the amorphous phase. Various effects could contribute to the
difference in crystallization temperature between clusters and the 10 nm thin
film, i.e. a composition effect, different surface to volume ratio, matrix influence,
stress or strain effects or an intrinsic size effect.
The third part of this thesis is dedicated to a study of the effect of interface
layers on the crystallization temperature of GeTe and GST thin film (10 to 100
nm thick). Despite its broad scientific and technological interest, this subject
has not been widely treated in literature up to now. First, measurements of the
crystallization temperature of GST and GeTe 100 nm thin films embedded in
three different materials (Ta, TiN and SiO2 ) through reflectivity measurements
are reported. It is observed that in both GeTe and GST interfaced with Ta
the crystallization temperature is around 20C higher than the one obtained by
interfacing those materials with TiN or SiO2 . Even if some studies in literature
put in evidence the influence of interface over the crystallization temperature of
Ge-Sb-Te thin films, such a remarkable interface effect in relatively thick films
3
(100 nm) was never reported before. The structural properties of the crystalline
phase of GeTe films, such as the grain size and texture, are investigated through
X-Ray diffraction analysis for the three different interface materials. The results
show that the SiO2 interfaced samples are characterized by a strong texture while
a weak texture is observed for Ta and TiN interfaced samples. Moreover, the
grain size calculated from the 012 Bragg peak width for SiO2 interfaced samples
is bigger than the ones calculated for Ta and TiN interfaces for a tilting angle
of 40of the sample. This suggest that for planes tilted by 40compared to
the (012) plane the SiO2 /GeTe surface is energetically favorable, resulting in
an abnormal growth with a preferred orientation with the (100) or (010) plane
parallel to the sample surface. If this hypothesis is true, a different nucleation
and growth mechanism for the different interfaced samples can be supposed.
In the last part, general conclusions and perspectives are presented.
4
Resume
Les memoires a changement de phase (PCM) sont l’un des candidats les plus
prometteurs pour la prochaine generation de memoires non-volatiles (NVM),
dont le marche est domine par la technologie Flash. Les PCM sont basees sur la
propriete des certains materiaux chalcogenures, appeles materiaux a changement
de phase (PC), de changer reversiblement d’etat entre une phase cristalline et une
phase amorphe. Ces deux phases sont caracterisees par des proprietes electriques
et optiques tres differentes, ce qui rend possible de stocker des informations.
Pour que les PCM soient competitives un grand effort de recherche est encore
necessaire. Cet effort devrait etre dedie d’une part a l’amelioration des aspects
technologiques, comme l’architecture des dispositifs, le layout et le circuit de
controle, et d’autre part a l’optimisation des materiaux PC utilises dans la cellule
de memoire. L’effet de la reduction de taille et celui des couches d’interface sur
les proprietes de materiaux PC doivent aussi etre compris. Jusqu’ici le materiau
PC le plus utilise et etudie est Ge2Sb2Te5 (GST), parfois dope avec de l’azote,
mais la recherche est extremement active dans l’investigation d’autres composes
qui peuvent offrir de meilleures proprietes. Les materiaux a utiliser dans les
PCM doivent cristalliser vite, montrer un grand contraste optique et electrique
entre les phases amorphes et cristallines, avoir une temperature de fusion suff-
isamment basse pour limiter l’energie electrique necessaire pour l’amorphisation
et la stabilite de leur phase amorphe doit etre grande pour garantir une bonne
retention des donnees. Le dopage des materiaux PC a ete identifie comme une
5
solution prometteuse pour l’optimisation de proprietes. De plus, ces proprietes
ne doivent pas se deteriorer apres beaucoup de cycles de transformation en-
tre les etats amorphes et cristallins et doivent etre conserves quand la taille est
reduite, c’est-a-dire pour une petite quantite de materiau PC incorpore entre des
materiaux conducteurs et/ou des isolants. Dans ce contexte, le but de ce travail
est d’examiner l’effet de la reduction de taille et du dopage sur les proprietes
des materiaux PC. Trois axes ont ete explores. Le premier est une contribu-
tion pour comprendre l’impact du dopage sur la structure de GeTe amorphe. Le
deuxieme est une investigation de l’effet de confinement en 3D sur le changement
de phase de GST. Le troisieme est l’etude d’influence du materiau d’interface
sur la cristallisation des films minces de GST et GeTe.
Dans la premiere partie de la these la structure locale de GeTe amorphe
dope avec C ou N est determinee par des experiences de diffusion des rayons X
executees au synchrotron SOLEIL (Saclay, France). Au debut de ce travail il
avait ete demontre que le dopage par C ou N de GeTe ameliore spectaculaire-
ment la retention de donnees et baisse le courant necessaire pour l’amorphisation
dans les dispositifs. Le but etait donc de comprendre l’impact du dopage
sur la structure amorphe de GeTe en analysant la fonction de distribution
de paires d’echantillons dopes et pas dopes. L’impact du dopage est revele
experimentalement par l’apparition d’un nouveau pic dans la fonction de dis-
tribution de paires de GeTe dope, indiquant la formation d’une liaison a une
nouvelle distance, absente dans le materiau amorphe non dope. Des simula-
tions ab initio montrent la formation de nouveaux environnements tetraedriques,
triangulaires ou pyramidaux centres sur le carbone ou l’azote ainsi que des
longues chaınes de carbone et quelque molecules d’N2. Le nouveau pic observe
experimentalement correspond aux distances Ge-Ge dans les unites centrees
sur le carbone ou l’azote. Ces changements structurels peuvent etre relies a
l’augmentation de la temperature de cristallisation et de l’energie d’activation
du GeTe dope C ou N.
L’effet de confinement sur le changement de phase de GST est le sujet de
la deuxieme partie de cette these. La capacite des materiaux PC a etre con-
fines en maintenant leurs proprietes est une exigence fondamentale pour le
6
developpement des PCM. Il y a eu beaucoup d’etudes sur des films minces
d’epaisseur variable. Dans quelques cas, une tres forte augmentation de la
temperature de cristallisation a ete trouvee pour des films plus minces que 10
nm, jusqu’au point de perdre la propriete de changement de phase pour des films
plus minces que 2 nm. Ces etudes traitent du confinement dans une seule direc-
tion, l’epaisseur de film, mais le systeme ideal pour etudier l’effet de la reduction
de taille sur des materiaux PC dans une cellule de memoire (ou tant la taille que
les effets d’interface jouent un role) est un ensemble de nanoparticules. En effet,
la taille des particules est limitee dans trois dimensions et les effets d’interface
sont augmentes en raison d’un plus grand rapport surface/volume. Des agregats
de GST, avec une taille moyenne d’autour de 5.7 nm (± 1 nm) et deposees
dans une matrice de Al2O3 , ont ete fabriques par pulverisation puis deposes et
caracterises par des mesures de diffraction des rayons X faites au synchrotron
ESRF (Grenoble). A l’heure actuelle, ces particules sont les plus petites partic-
ules de GST jamais deposees par pulverisation et avec la distribution de taille la
plus etroite. Ces particules ont ete fabriquees par une methode proche de celle
utilisee pour la deposition de films minces dans les dispositifs PCM, et donc
les informations obtenues peuvent etre facilement exportees vers la fabrication
des dispositifs. La temperature de cristallisation des nanoparticules est autour
de 180C, seulement legerement au-dessus de la temperature de cristallisation
(155C ) d’un film mince de GST de 10 nm depose dans les memes conditions et
encapsulee par Al2O3 . La phase cristalline est la phase metastable cubique de
GST tant pour les nanoparticules que pour le film mince. Le parametre de maille
est plus grand dans les nanoparticules que dans le film mince. On peut expliquer
la variation de parametre de maille en supposent que la matrice Al2O3 , qui est
rigide, exerce une tension sur les nanoparticules et que leur volume pendant la
cristallisation est force de rester constant et egal au volume occupe dans la phase
amorphe. Divers effets pourraient contribuer a la difference entre la temperature
de cristallisation des nanoparticules et du film mince de 10 nm, soit un effet de
composition, la difference de rapport surface/volume, l’influence de la matrice,
des effets de contraints et deformations ou un intrinseque effet de taille.
La troisieme partie de cette these est consacree a une etude de l’effet de
7
couches d’interface sur la temperature de cristallisation de films minces de GeTe
et GST (epaisseur de 10 a 100 nm). Malgre son large interet scientifique et
technologique, ce sujet n’a pas ete largement traite dans la litterature jusqu’ici.
D’abord, la temperature de cristallisation de films minces de GST et GeTe de 100
nm encapsules dans trois materiaux differents (Ta, TiN et SiO2 ) est determinee
par des mesures de reflectivite. Il est observe que, tant dans le GeTe que dans le
GST, la temperature de cristallisation obtenue dans le cas d’interface avec du Ta
est autour 20C plus haute que celle obtenue en interfaant ces materiaux avec
TiN ou SiO2 . Meme si quelques etudes de la litterature ont mis en evidence
l’influence d’interfaces sur la temperature de cristallisation de films minces de
Ge-Sb-Te, un effet d’interface si remarquable dans des films relativement epais
(100 nm) n’a jamais ete rapporte auparavant. Les proprietes structurelles de
la phase cristalline de films du GeTe, comme la taille de grains et la texture,
sont examinees par diffraction des rayons X pour les trois differents materiaux
d’interface. Les resultats montrent que l’echantillon interface avec SiO2 est car-
acterise par une texture forte tandis qu’une texture faible est observee pour les
echantillons interfaces avec Ta et TiN. De plus, la taille de grains, calculee en util-
isant la largeur du pic de Bragg 012, pour l’echantillon interface avec SiO2 est
plus grande que celle calculees pour les interfaces Ta et TiN pour un angle
d’inclinaison de l’echantillon de 40. Ceci suggere que l’interface SiO2 /GeTe
est energetiquement favorable pour les plans inclines de 40par rapport aux
plans 012, aboutissant a une croissance anormale dans cette direction. Si cette
hypothese est vraie, on peut supposer un mecanisme de nucleation et crois-
sance different pour les echantillons interfaces avec differents materiaux. Dans
la derniere partie, des conclusions generales et des perspectives sont presentees.
8
Chapter 1
Phase Change Memories
Overview
9
1.1 Introduction
1.1 Introduction
The necessity to store informations has always been one of the basic needs of
mankind throughout its history. In the last few decades, with the explosive de-
veloping of electronics and computers, this need became an urgency. Since the
formulation of Moore’s Law in 1965 [1] the microelectronics industry develop-
ment has been ruled by the trend of reducing device cost of one half every two
years. This has been achieved by conventional CMOS device architectures by
increasing the integration density of devices, meaning reducing their dimensions
or even developing new strategies as 3D device integration that allows higher
density at the same device size. Floating gate non-volatile memories (NVM),
usually named Flash memories, represent the mainstream in the NVM market.
They have been the reference technology for years, but their further scaling be-
come difficult due to technological and physical constraints. Already over the
past few years, Flash memories have faced hard challenges for keeping the scal-
ing trend and great difficulties arise for the next technology nodes that make
hard to even maintain actual specifications.
As a consequence, there is a rising industrial interest for emerging alterna-
tive NVM technologies that can offer better scaling possibilities with even better
memory performances than Flash memories. Among those alternatives the most
interesting one is the Phase-Change resistive memory (PCM). PCM are based on
the properties of some chalcogenide materials, called phase change (PC) mate-
rials, to reversibly switch between a crystalline and an amorphous phase. These
two phases are characterized by very different electrical and optical properties
and this makes possible to store the information in terms of ’0’ and ’1’ levels,
as it will be described in details in section 1.2. Phase-change memories, also
called Ovonic Unified Memory, offer a scalability beyond Flash technology and
can potentially be better than Flash memories in terms of faster random access
time, higher density and higher endurance [2].
The first to investigate extensively the switching properties of chalcogenides
and to demonstrate their practicality was S. R. Ovshinsky, during the 1950s and
1960s [3]. In early 1970s the interest in PCM rose and a first 256-bit memory
10
1.1 Introduction
array of phase change memory cells was developed by R. G. Neale, D. L. Nelson
and G. E. Moore in 1970. However, the programming operation parameters of
this memory were very poor. Other studies on phase change memory electrical
devices have been done until 1978 but the enormous power required for pro-
gramming constituted a strong limitation and they were abandoned (Chapter
1 of Ref. [4]). The reason for this enormous power consumption were the big
dimensions of the memory cell. The PCM devices were not as competitive as the
metal oxide semiconductor memories that were rising in market at those times,
so the interest in PCM subsided and there were no further developments until
the first years of 2000.
However, during the 1970s and 1980s chalcogenide materials continued to
attract interest because of their suitability for use in optical memories. The
most important result of this effort was the discovery in 1987, by Yamada and
co-workers, of the GeTe − Sb2Te3 pseudobinary line [5] and in particular of
the Ge2Sb2Te5 compound, called GST [6]. The properties of those materials
opened the way to the development of phase change optical recording devices
such as Compact Disk (CD), Rewritable Compact Disk (CD-RW), Digital Ver-
satile Disk (DVD) and Blue-Ray DVD. On the other hand, at the beginning of
2000s the new shrinking possibilities obtained through the great development
of lithographic techniques triggered a new interest in phase change memory
electrical devices [4]. The now reduced available volume of the memory cell,
resulting in lower power consumption, led PCMs to be developed by the Re-
search and Development departments of many industrial companies as Intel,
STMicroelectronics and Samsung. Nowadays, PCMs are one of the best can-
didates to replace market-leader Flash Non Volatile Memories (NVM) that are
approaching their critical size limit.
One of the open research subjects in the PCM field is material study and
optimization. While the device architecture faces the challenges of scaling and
vertical integration [2], new materials that can offer better performances should
be investigated. A deeper understanding of the phase change mechanism and
how PC materials structures are related to their properties is mandatory. Nowa-
days, this need is combined with the rising scientific interest of Phase Change
11
1.2 Phase Change Memories working principle
Figure 1.1: Basic principle of the phase transformation [8]. PC materials can switch re-
versibly between an amorphous state, corresponding to the logical level ’0’ or RESET, and a
crystalline state corresponding to the logical level ’1’ or SET. The SET operation consists in
programming the cell into the SET state, while the RESET operation consists in programming
the cell into the RESET state. To obtain the amorphous phase the PC material must be an-
nealed above its melting temperature and then rapidly cooled down. To obtain the crystalline
phase the material must be annealed above its crystallization temperature Tx .
Materials. In the last years the importance of the basic research on PC mate-
rials has been confirmed by an increasing interest of the scientific community,
and by the success of dedicated symposium as, for example, the Phase Change
symposium in the Material Research Society Spring conference. The number of
articles dedicated to studies of the fundamental properties of PC materials has
rose in the last years (see for example Ref. [7] and [8]), confirming that this is
an active and promising research field.
The aim of this thesis is to study those materials, focusing on the effect of
doping well known compounds, the effect of different material interfaces and the
shrinking effect on very small nanoparticles. The following of this first Chapter
will be dedicated to an introduction on PCM basic working principle (Section
1.2), the description of the theory of amorphization and crystallization (Section
1.3) and a brief review of the main structural characteristics of PC materials,
focusing on the structure of the well known compounds GeTe and GST (Section
1.4).
1.2 Phase Change Memories working principle
Phase change memories (PCM) are based on the property of so called phase
change materials (PC materials) to change reversibly between an amorphous
and a crystalline state, as schematically shown in Figure 1.1.
12
1.2 Phase Change Memories working principle
The phase change is obtained through heating of the material. In memory
devices the heating can be provided by electric or laser pulses. If a PC material
is heated above its melting temperature Tm and cooled down quickly (with a
cooling rate of 109 − 1011K/s) it solidifies in a glassy structure, the amorphous
phase. The glass is in a metastable state so it will tend to crystallize on very
long time scale. This time should be of the order of several years since it deter-
mines the capability to maintain the information. If the amorphous material is
annealed for a sufficiently long time (usually tens of nanoseconds) below Tm but
above its crystallization temperature Tx , it switches to the crystalline phase.
The theory of the phase change mechanism will be explained with more details
later in Section 1.3. The two phases are characterized by very different optical
and electrical parameters, thus providing the contrast required to distinguish
between logical states. For example, the amorphous phase exhibits a high value
of resistivity and a low value of reflectivity, and vice-versa for the crystalline
phase. The optical contrast between the amorphous and crystalline phase is
illustrated in Figure 1.2 for the case of Ge2Sb2Te5 .
The phase change property allows to store an information by associating the
logical level ’0’ and ’1’ to the two different phases. Traditionally, the level ’0’
(or RESET level) has been associated to the amorphous phase and the level
’1’ (or SET level) to the crystalline phase. Crystallization is the slowest process
involved. It must occur quickly in the programming operation in order to achieve
fast programming speed, but the spontaneous crystallization of the metastable
amorphous phase should not take place for many years at room temperature in
order to grant data retention. This means that the crystallization rate of PC
materials must increase by orders of magnitude with the change in temperature
between room temperature and Tx .
In the following sections the working principle of a schematic cell device will
be explained and the programming curves and electrical characteristics of the
cell will be presented.
13
1.2 Phase Change Memories working principle
Figure 1.2: Evidence of the different optical properties of the PC material Ge2Sb2Te5 in the
amorphous and crystalline phases. The reflectivity of GST is reported as a function of tem-
perature starting from an initially amorphous sample. The amorphous phase is characterized
by a low reflectivity value compared to the one of the crystalline phase. On the graph it is
easy to identify the crystallization temperature at which the phase transformation occurs.
14
1.2 Phase Change Memories working principle
Figure 1.3: Schematic representation of the lance-like structure of a PCM cell device. The
PC material is interfaced with a top electrode and a bottom electrode (heater).
1.2.1 Basic device example
The schematic picture of a Ovonic Unified Memory (OUM) PCM cell in its
simplest form (lance-like structure) is reported in Figure 1.3.
The device consists in a thin film of PC material which is electrically acces-
sible by a top electrode and a bottom electrode, also called heater . For almost
all the PC materials integrated in devices the as-fabricated cell is entirely crys-
talline, due to the high temperatures reached in fabrication process. In order
to read and program the cell, an imposed external voltage is applied at the
electrodes generating a current that flows from to the top electrode the heater
through the PC material. To read the state of the cell a low power current pulse
is imposed and the overall resistance of the cell is measured. If the resistance is
high the cell is in the RESET state, while if the resistance is low the cell is in
the SET state. It is worth noting that the cell is considered to be in the SET
state when the read resistance value is sufficiently low, and this can be achieved
by the formation of crystalline paths percolating through the amorphous volume
and not necessarily by crystallizing the entire volume. Concerning the program
operations, the shape and intensity of the applied current pulse determines if
the cell will be programmed in the amorphous state (RESET operation) or in
15
1.2 Phase Change Memories working principle
Figure 1.4: Current pulses for the programming operation of the cell. RESET pulse (a)
SETMIN pulse (b) and SET pulse (c).
the crystalline state (SET or SETMIN operation) as shown in Figure 1.4. The
contact area between the PC material and the heater is very narrow so the
current density reaches its highest value at the PC material - heater interface.
As shown in Figure 1.4a, for the RESET operation a current pulse of high in-
tensity with a rapid falling edge is applied (RESET pulse) so that the current
density at the PC material / heater interface is sufficiently high to heat the
PC material over the melting temperature by Joule heating. The abrupt falling
edge of the RESET pulse induces a fast quench of the material that solidifies in
the amorphous phase in an hemispherical volume. Amorphization of this area
blocks the low-resistive current path and results in an overall large resistance.
It is worth underlining that reducing the heater dimension, meaning reducing
the cell size, results in an increased current density so that the current required
for amorphization is reduced. The SET pulse can be chosen to have the same
shape as the RESET pulse but with a lower intensity and a longer duration
(SET pulse, Figure 1.4b) or an intensity sufficiently high to melt the PC mate-
rial but with a slow falling edge (SETMIN pulse, Figure 1.4c). In the first case
the PC material is heated below its melting temperature and the amorphous to
crystalline transition is induced. Usually the transformation does not involve
all the amorphous volume but results in the creation of percolating paths, as
explained before. Thus the overall resistance is higher than the resistance of an
entirely crystalline cell, but still two or three order of magnitude lower than the
RESET value. In the SETMIN case the PC material is first melted and then
crystallized by a slow quenching and the final resistance value is lower than in
the SET case.
16
1.2 Phase Change Memories working principle
Figure 1.5: I-V characteristic of a PCM cell in the crystalline and amorphous states (from
Ref.[9]). The I-V characteristic of the amorphous state present a snap-back in correspondence
of a threshold voltage that is not present in the crystalline I-V curve.
1.2.2 Electrical conduction model
The mechanism of electrical conduction in phase change memories is a very ac-
tive research field, due to the important role of electrical properties in devices.
The current-tension (I-V) characteristic of a PCM cell is reported in Figure 1.5.
It is possible to distinguish between a low electrical field region and an high elec-
trical field region. For low applied voltages (low electrical field region) the GST
amorphous conductance is low compared to the crystalline GST conductance.
When the external bias reaches a certain value (called threshold switching volt-
age) a snap-back takes place and the conductance abruptly switches to a higher
value. This threshold switching, also called Ovonic Threshold Switching (OTS),
was first discovered by Ovshinsky [3]. The crystalline GST I-V curve presents
no evidence of such a switching and in the high field region the conductances
of both states are equal. It is important to underline that the threshold switch-
ing does not correspond to the amorphous to crystalline transition. After the
switching takes place, the cell remains amorphous until Joule heating is suffi-
cient for inducing crystallization. The OTS is fundamental in order to grant
low power dissipation during the SET operation. When the voltage applied to
the amorphous cell exceeds the threshold value, the current flowing through it
17
1.3 Physics of phase change transformations
increases drastically, allowing the phase transition to take place with a low ap-
plied power. Without the switching phenomenon, the high power required to
perform the SET operation would make the programming operation unpracti-
cal. The physical mechanism for OTS has been widely investigated. Even if it is
still not conclusively clarified, it is generally supposed to be an electronic effect
rather than a thermal or structural effect. Ielmini and Zhang proposed a model
for conduction [10, 11] described by the Poole-Frenkel effect that well describes
the transport properties of the crystalline and amorphous phases as well as the
switching effect.
1.3 Physics of phase change transformations
1.3.1 Amorphization
As already stated in the introduction, the amorphous phase can be obtained from
the melt by a rapid quench, fast enough to avoid crystallization. If a liquid is
cooled below its melting temperature Tm it does not crystallize instantaneously
and it can be undercooled. While the temperature decreases, the liquid viscosity
η increases. Such an undercooled liquid is in a metastable equilibrium, meaning
that it is metastable with respect to the crystalline stable phase but it is still
in its internal equilibrium. If the undercooled liquid is cooled down below the
so-called glass transition temperature Tg it becomes configurationally frozen,
losing the thermal equilibrium, and it becomes a glass. Tg is commonly defined
as the temperature at which the time scale necessary for atomic rearrangements
becomes larger than the measurement time, and it usually occurs at the point
where the viscosity equals 1 × 1012 Pa s [12, 8]. Below Tg the glass has a very
low microscopic atomic mobility D(T ), which is inversely proportional to the
macroscopic viscosity η(T ) according to the Stokes-Einstein relation D(T ) ∝
T/η(T ).
Crystallization is thermodynamically forbidden above Tm and it is extremely
slow below Tg , while it can rapidly occur for temperatures between Tg and
Tm. For temperatures slightly below Tm the driving force for crystallization
is so low that crystallization would occur only very slowly, so with a rapid
18
1.3 Physics of phase change transformations
Figure 1.6: Time-temperature-transformation (TTT) diagram for a PC material taken from
Reference [8]. The phase transformation of a fixed volume of PC material is reported depending
on the time spent at a certain temperature. The two orange lines on the graph indicate two
different constant rate quenching processes while the two purple lines indicate two annealing
processes starting at room temperature.
quenching of the liquid below Tg it is possible to avoid crystallization and result
in the formation of the amorphous phase. This is illustrated in Figure 1.6a,
reported from Ref. [8], in which the phase transformation is shown in terms of a
time-temperature-transformation (TTT) diagram. The corresponding mobility
and the driving force for crystallization are reported in Fig. 1.6b and 1.6c,
respectively. In the TTT diagram the phase transformation of a fixed volume of
PC material is reported, depending on the time spent at a certain temperature.
The two orange lines on the graph indicate two different constant rate quenching
processes. If the cooling is sufficiently fast (around 109K/s) crystallization can
be avoided slightly below Tm due to the very low driving force for crystallization
(Figure 1.6c). With further fast undercooling, the driving force increases but
the mobility decreases and if Tg is reached the material becomes amorphous.
On the other hand, if the cooling rate is too slow the material will crystallize.
The amorphous phase can crystallize below Tg only for very long times. This
time has a very important role in determining the data retention in a memory
cell, as it defines the time at which a cell in the RESET state will crystallize
losing the stored information. On the TTT graph are also reported in purple two
annealing processes starting at room temperature, showing that crystallization
time and temperature are dependent.
19
1.3 Physics of phase change transformations
1.3.2 Crystallization
Two different process contributes to crystallization of an amorphous solid.The
first one is the nucleation, that initiates the crystallization through the formation
of small crystalline nuclei. The second one is the growth of those nuclei to
a macroscopic size. The so-called classical nucleation theory of steady state
nucleation has been developed by Volmer, Weber, Becker, Doring, Turnbull and
Fisher during the first decades of the 20th century [13, 14, 15, 16], based on the
pioneering work of Gibbs [17].
1.3.3 Nucleation
Nucleation can occur in two different ways. In the first and simplest case, called
homogeneous nucleation, the crystallite germinates inside the amorphous phase,
without involving other substances. If instead the amorphous phase is in contact
with other substances that act as preferred sites for nucleation, an heterogeneous
nucleation occurs.
Homogeneous Nucleation
Form Gibbs’ thermodynamical theory, clusters of radius r can be formed inside
the amorphous phase by thermodynamics fluctuations. Their equilibrium distri-
bution is ruled by the Boltzmann statistic, meaning that the number of clusters
of radius r at equilibrium is
N equ(r) = N0 · exp(−∆Gcluster(r)
kBT) (1.1)
where N0 is the total number of atoms in the liquid, ∆Gcluster(r) is the reversible
work for crystal cluster formation, kB is the Boltzmann constant and T is the
absolute temperature. ∆Gcluster(r) can be written as
∆Gcluster(r) = −∆Glc,V ·4
3πr3 + 4σπr2 (1.2)
where ∆Glc,V is the Gibbs free energy difference per volume between the crys-
talline and the amorphous phase and σ > 0 is the interfacial free energy. ∆Glc,V
is zero at Tm and positive for T <Tm . So Eq. 1.2 is composed of a nega-
tive volume term that becomes larger as the temperature T is reduced below
20
1.3 Physics of phase change transformations
Figure 1.7: Evolution of ∆Gcluster(r) as a function of r corresponding to Eq. 1.2, taken
from Chapter 7 of Reference [4]. The curve exhibit a maximum for the r = rc (critical radius)
that corresponds to the critical work for cluster formation ∆Gc.
Tm and a surface term, always positive, that results from the creation of a clus-
ter/liquid interface. The evolution of ∆Gcluster(r) as a function of r is depicted
qualitatively in Figure 1.7.
The curve present a maximum at
rc =2σ
∆Glc,V
(1.3)
where the critical radius rc is of the order of a few nanometers. The energy
of a nucleus of radius rc may be calculated by substituting rc to r in 1.3, thus
obtaining the critical work for cluster formation ∆Gc
∆Gc = ∆Gcluster(rc) =16π
3
σ3
(∆Glc,V )2. (1.4)
Clusters with radius r = rc are the so-called critical clusters. The evolution
to bigger dimensions of clusters with r < rc is energetically not favorable so
they spontaneously decay, while clusters with r > rc can grow due to the free
energy gain. This means that ∆Gc constitutes a barrier against crystallization,
the same barrier that impedes immediate crystallization of the amorphous phase
when it is undercooled below Tm .
The approach of Gibbs is purely thermodynamic, and on its basis a first
kinetic model for nucleation was developed by Volmer and Weber [13, 14]. They
21
1.3 Physics of phase change transformations
modified Eq. 1.1 by taking into account the fact that it becomes unphysical
for r > rc, where the clusters distribution at equilibrium begin to increase with
increasing r. To avoid this, N equ(r) was set to zero for r > rc. By considering
that the nucleation occurs when a critical cluster gains one more atom, the
nucleation rate was calculated per unit volume per second
Iequ = sc · k ·N equ(rc) = sc · k ·N0 · exp(−∆Gc
kBT) (1.5)
where k is the arrival rate to the crystalline cluster of amorphous phase atoms
and sc is the number of surface atoms of the cluster.
One of the limitation of the Volmer-Weber theory is that a critical cluster
that gain one more atom is supposed to grow to macroscopic size while in reality
there is still a probability for it to decay. Backer and Doring [13, 14, 15] proposed
a different expression for the equilibrium cluster distribution N equ(r) that takes
into account that possibility, thus obtaining the following steady state nucleation
rate
Iss = sc · k ·N0 ·1
ic· (
∆Gc
3πkBT)
︸ ︷︷ ︸
ΓZ
·exp(−∆Gc
kBT) = Iequ · ΓZ (1.6)
where ic is the number of atoms in the critical clusters and ΓZ is the so-called
Zeldovich factor, which has been found to be usually between 1/100 and 1/10.
The weak dependence of ΓZ on temperature, especially if compared with the
exponential term, makes Eq. 1.6 essentially equal to Eq. 1.5 for most practical
purposes, but in the case of Eq. 1.6 the kinetic problem has been correctly
treated.
Up to the Volmer-Weber model, all the results were obtained by considering
a gas as a amorphous phase. In this case the value of the arrival rate k was calcu-
lated from the gases theory. The value of the pre-exponential factor of 1.6 for an
amorphous material was calculated by Turnbull and Fisher [16], who completed
the classical nucleation theory. They distinguished between a diffusion-limited
crystallization and a collision-limited crystallization. The former is the case of
phase change materials, and the expression for k is
k =6D
λ2(1.7)
22
1.3 Physics of phase change transformations
where λ is the average interatomic distance. By using the Stokes-Einstein
equation that relates the diffusion coefficient D and the viscosity η it is possible
to write
ηD =kBT
3πλ, (1.8)
and the nucleation rate Iss for the diffusion-limited model can be expressed as
a function of η:
Iss = sc ·2kBT
ηπλ3·N0 · ΓZ · exp(−
∆Gc
kBT) (1.9)
The pre-exponential factors of Iss can be estimated in both cases by sub-
stituting reasonable values for N0, sc T and ΓZ . They result to be 1036
ηfor the
diffusion-limited case and 1039 for the collision-limited case, with an uncertainty
of two to four orders of magnitude. This has not a great influence on the overall
expression due to the strong dependence of Iss on ∆Gc and σ, both present
in the exponential term. The diffusion-limited Iss tends to zero near Tm and
Tg and exhibit a maximum for a temperature intermediate between them, as it
happens for phase change materials. This is not the case for collision-limited
kinetics, where Iss increases continuously as the temperature decreases below
Tm .
Heterogeneous Nucleation
The model for heterogeneous nucleation was developed by Volmer and Weber
[18]. It is based on the Gibbs’ model already described for the homogeneous nu-
cleation, but considering a flat substrate that act as a heterogeneous nucleation
site.
In Figure 1.8 is reported the model for heterogeneous nucleation, taken from
Chapter 7 of Reference [4]. In this model the crystalline cluster nucleates on the
heterogeneous substrate as a spherical cap of radius r. This spherical cap can
be considered as the exposed part of a complete sphere of radius r, so that the
fraction of the exposed volume can be calculated as a function of the wetting
angle θ
f(θ) =(2 + cosθ)(1− cosθ)2
4. (1.10)
23
1.3 Physics of phase change transformations
Figure 1.8: Model for the heterogeneous nucleation taken from Chapter 7 of Reference [4].
The crystalline cluster is a spherical cap which correspond to the exposed part of a sphere
of radius r. In the schematic picture are also reported the wetting angle θ and the crystal-
substrate, amorphous-substrate and amorphous-crystal interfacial energies (respectively σcs,
σls and σlc).
It was demonstrated by Volmer and Weber that the free energy for cluster
formation ∆Gcluster is reduced if
σcs − σls < σlc (1.11)
where σcs, σls and σlc are the crystal-substrate, parental phase-substrate and
parental phase-crystal interfacial energies, respectively. In this case, ∆Gcluster
for the heterogeneous nucleation is the ∆Gcluster for the homogeneous nucleation
multiplied by the factor f(θ)
∆Ghetc = ∆Ghet
c · f(θ). (1.12)
The critical radius rc remains unchanged, and the whole model for homo-
geneous nucleation is still valid with the only difference of a lower ∆Gcluster.
However, the parent phase atoms that can act as nucleation sites are not all the
atoms in the parent phase but only those interfaced with the substrate, so their
number is decreased. If ǫ is the fraction of parent phase atoms that are in con-
tact with the substrate on the total, the ratio of homogeneous and heterogeneous
24
1.3 Physics of phase change transformations
nucleation rates is
Iss,het
Iss,hom= ǫ · exp(
∆Gc
kBT· [1− f(θ)]). (1.13)
The heterogeneous nucleation rates have been observed to be much higher than
the homogeneous ones, implying that θ must be small.
1.3.4 Growth
When a cluster has reached its critical radius it grows to a macroscopic size.
This growth is interface-controlled by the addition of new parental phase atoms
in the crystalline cluster [13]. The crystal growth velocity is
u = γs · λ · k · [1− exp(−∆Glc,atom(T )
kBT)](T < Tm) (1.14)
where 0 < γs < 1 is the fraction of sites on the interface where a new parent phase
atom can be incorporated, λ is the average interatomic distance, ∆Glc,atom(T )
is the Gibbs free energy between the parent phase and the crystalline phase per
atom and k has the same meaning and value as for nucleation. For diffusion-
limited kinetics, by substituting k as in Section 1.3.3 and using Eq. 1.8 the
crystal growth velocity is
u = γs ·2kBT
ηπλ2· [1− exp(−
∆Glc,atom(T )
kBT)](T < Tm) (1.15)
As for the nucleation rate, the growth velocity u is zero at T =Tm , negligible
at T = Tg and exhibit a maximum in the temperature range between Tg and
Tm .The growth rate maximum is usually located at a higher temperature than
the nucleation rate maximum.
1.3.5 Johnson - Mehl - Avrami - Kolmogorov (JMAK)
formalism
The so-called JMAK (Johnson-Mehl-Avrami-Kolmogorov) model is an alterna-
tive to the classical crystallization theory for describing the crystallization ki-
netics. While the classical nucleation theory allows the calculation of the cluster
nucleation and growth rates, as well as their size distribution, the JMAK model
25
1.3 Physics of phase change transformations
is a mean to calculate the crystalline fraction in terms of crystal nucleation and
growth rates.
The model is based on the JMAK equation, which gives the volume frac-
tion of the transformed material as a function of time (x(t)) under isothermal
annealing conditions:
x(t) = 1− exp (−ktn) (1.16)
where t is time, k is an effective rate constant and n is the so called Avrami coef-
ficient. The value of k depends on temperature through the Arrhenius equation
k(T ) = νexp
(EA
−kBT
)
(1.17)
where ν is the frequency factor, EA is the activation energy, T is the absolute
temperature and kB is the Boltzmann constant. The value of ln [−ln (1− x)]
plotted as a function ln(t) is the so-called JMAK plot. In literature, the JMAK
theory has been often used to interpret isothermal annealing of PC materials,
and the activation energy EA has been usually determined through Kissinger
analysis [19]. The Kissinger method is based on the measurement of the variation
of the crystallization temperature Tx for different heating rates dT/dt, which are
related to the activation energy through the equation
ln
(1
T 2x
·dT
dt
)
= −EA
kBTx
+ C (1.18)
so that it is possible to deduce the activation energy as the slope of the linear in-
terpolation of the plot of ln(
1T 2x· dT
dt
)
versus 1kBTx
. However, even if this method
is widely used, it is based on the assumption of an Arrhenius-like temperature
dependence for crystallization. This is not the case when the crystallization is
controlled by the nucleation rate, which is non-Arrhenius [20].
Eq. 1.16 can be applied under the conditions that nucleation occurs ran-
domly and uniformly with a time independent rate and that growth is interfaced-
controlled and with a size independent rate. The use of Eq. 1.16 should not
be legitimate without fulfilling those conditions, which are usually not verified
for GST, and as a consequence the values reported in literature for EA and
ν differ significantly one from each other. However, due to its simplicity the
Kissinger method will be used in Chapter 3 and 4 of this thesis for quantitative
comparisons between materials.
26
1.4 Phase change materials
Required property of PC material Specification
High-speed phase transition Induced by nanosecond laser or voltage pulse
Long thermal stability of amorphous state At least several decades at room temperature
Large optical change between the two states Considerable difference in refractive index or ab-
sorption coefficient
Large resistance change between the states Natural consequence of the phase transformation
Large cycle number of reversible transitions More than 100,000 cycles with stable composition
High chemical stability High water-resistivity
Table 1.1: Properties that characterize PC materials [7].
1.4 Phase change materials
Materials to be used in PCM should meet several strict requirements. They
should possess the properties of a fast crystallization, a large optical and elec-
trical contrast between the amorphous and crystalline phases, a melting tem-
perature sufficiently low to limit the electric power needed for amorphization
and a high stability of the amorphous phase to grant good retention perfor-
mances. Moreover, those properties must not decay with cycling between the
states. The crucial properties of phase-change alloys are summarized in Table
1.1, taken from Reference [7].
1.4.1 Ge:Sb:Te compounds
As already described in Section 1.1, from the material point of view the greatest
discovery for phase change memories was done in the 1980s by Yamada and
his coworkers. They identified the materials belonging to the GeTe − Sb2Te3
pseudo-binary line in the Ge:Sb:Te ternary phase diagram as the ones with the
best properties. In Figure 1.9 the Ge:Sb:Te phase diagram is represented, with
several PC materials and the GeTe− Sb2Te3 pseudo-binary line. In particular,
the Ge2Sb2Te5 compound (simply called GST) became a standard for optical
storage devices that have been developed during the 1990s. It was chosen for
its high retention time, fast transformation speed and large optical contrast be-
tween crystalline and amorphous phase. Thus, GST was chosen as the active
27
1.4 Phase change materials
Figure 1.9: PC materials reported on the ternary Ge:Sb:Te phase diagram, with the GeTe−
Sb2Te3 pseudo-binary line put in evidence (taken from Ref. [8]).
material to be first employed in PCM due to the wide number of studies already
performed on it. However, the study of alternative phase change materials that
can offer better properties for PCM is very active and rich, both for its scien-
tific interest and the industrial development. For example, in 2011 Cheng and
coworkers studied compounds along the GeTe−Sb line. They demonstrated that
Ge-rich Ge2Sb1Te2 (the so-called golden composition) can offer better proper-
ties than GST in terms of crystallization speed and data retention [21]. It has
been also recently shown that GeTe compound can be a good candidate for em-
bedded PCM due to its higher crystallization temperature and better retention
time compared to GST [22]. The main properties of Ge2Sb2Te5 and GeTe are
compared in Table1.2
Phase change materials that belong to the GeTe − Sb2Te3 pseudo-binary
line are characterized by a few typical structural motifs indicating a common
bonding mechanism that could account for their properties. In general, ternary
compounds exhibit a rocksalt-like structure in their crystalline phase where the
anion sublattice is occupied by atoms of Te and the cation sublattice is randomly
occupied by Ge, Sb or vacancies [7].
The constant feature for all crystalline Ge:Sb:Te phase change alloys is the
presence of a more or less distorted octahedral-like coordination resulting from
a Peierls distortion. In addition, some materials are characterized by a consid-
28
1.4 Phase change materials
Properties Ge2Sb2Te5 GeTe
Crystallization temperature Tx [22] 145 C 185C
Activation energy EA (thin films) [23] 2.3 eV 2.0 eV
Activation energy EA (devices) [24] 3.13 eV 3.2 eV
Crystalline phase [25] Rocksalt cubic (with vacancies) Rhombohedral
Lattice parameters [25] a = 6.01 A a = 4.16 A, c = 10.69 A
Density (amorphous phase) [25] 5.86 g/cm3 5.60 g/cm3
Density (crystalline phase) [25] 6.13 g/cm3 6.06 g/cm3
Table 1.2: Comparison between the main properties of Ge2Sb2Te5 (GST) and GeTe.
erable number of vacancies (for example, 20% for Ge2Sb2Te5 [25] and 25% for
Ge1Sb2Te4 [7]). The amorphous phase differs considerably from the crystalline
one by the absence of a long range order, but a local order still exists in the amor-
phous structure of PC materials. It has been investigated both experimentally
and through ab-initio calculations.
In the following section the amorphous and crystalline structure of GeTe and
GST compounds will be described.
1.4.2 Structure of crystalline and amorphous Ge2Sb2Te5 and
GeTe
Both GeTe and GST are characterized by two different crystalline phases.When
amorphous Ge2Sb2Te5 (GST) is heated, it crystallizes at around 150C in a
metastable crystalline phase with a fcc rocksalt structure. The crystalline struc-
ture of Ge2Sb2Te5 (GST) has been described in 2000 by Yamada and coworkers
[26] as it is shown in Figure 1.10. The structure is a NaCl rocksalt structure with
an octahedral-like atomic arrangement, where the Te atoms occupy one lattice
site and the Ge and Sb atoms randomly occupy the second lattice site. The Te
sites are all fully occupied, while the Ge/Sb sublattice is characterized by the
presence of around a 20% of vacancies so that the structure is characterized by
local distortions. If cubic GST is heated above around 200C it transforms in
an hexagonal phase [6].
29
1.4 Phase change materials
Figure 1.10: Structure of GST in its crystalline metastable phase. One sublattice is occupied
by Te atoms (light blue) while the other is randomly occupied by Ge or Sb atoms (dark blue)
or vacancies (around 20% ). The cubic lattice parameter is 6.03 A [26].
30
1.4 Phase change materials
Figure 1.11: Structure of crystalline GeTe in its rhombohedral phase. The structure can be
described as a rocksalt-like structure, distorted by a relative shift of the sublattices along the
[111] direction. It is characterized by long (3.127 A) and short (2.87 A) Ge-Te bonds shown
respectively in white and green.
Amorphous GeTe crystallizes at 180C into a rhombohedral phase (space
group R3m) that is the stable crystalline phase at room temperature [25]. Above
around 430C this phase transforms into a cubic fcc rocksalt structure (space
group Fm3m) where Ge occupies one sublattice and Te the other sublattice. The
low temperature rhombohedral phase can be described as a slightly distorted
rocksalt structure obtained by a relative shift of both sublattices along the [111]
direction, so that each atom form three short bonds and three long bonds with
its nearest neighbors. The structure is represented in In Fig.1.10, and it can be
easily observed that each atom is in a distorted octahedral environment.
The structure of the amorphous GeTe and GST phases have been widely
studied in literature [27, 28] but in the following the subject will be treated
only briefly. The short and medium range order has been studied in litera-
ture mainly through X-ray scattering and measurement of the pair distribution
function method (see Chapter 2) that allow to determine the local atomic en-
vironments. It has been observed that a local chemical ordering takes place in
the amorphous phase by an alternation of Te and Ge(or Sb) atoms which is a
31
1.5 Goals and outline
precursor of the order in the crystalline phase. The first Te-Ge distance has been
confirmed to be around 2.6A, but the Ge coordination number is still debated.
Some studies through extended X-ray absorption fine structure (EXAFS) and
X-ray absorption near-edge spectroscopy (XANES) conclude that germanium is
tetrahedrally coordinated [29]. They were further supported by X-ray fluores-
cence and Raman scattering experiments [30, 31, 32]. However, further studies
with X-ray diffraction data and reverse Monte Carlo simulations indicate for
GST bond angles around 90and no homopolar bonds, while only for GeTe the
presence of both homopolar Ge-Ge bond and a deviation in bond angles from
90was observed [27, 28, 33]. In conclusion, the chemical alternation of Te and
Ge(Sb) atoms is confirmed and germanium is found to coexist in both tetrahe-
dral and distorted octahedral environments.
For both GeTe and GST, the density of the crystalline phase is higher than
the one of the amorphous phase, so the volume is reduced after crystallization,
and the density change for GeTe is relatively higher than for GST. Amorphous
GeTe has a density of 5.60 g/cm3, while for the crystalline GeTe it is 6.06 g/cm3.
The density of amorphous GST is 5.86 g/cm3 and the one of crystalline GST is
6.13 g/cm3[25].
1.5 Goals and outline
For PCM to be competitive it is fundamental to develop memory devices with
good retention properties, high cyclability, high transformation speed, low power
consumption and most of all high integration density, meaning cells of reduced
dimensions. Those requirements can be fulfilled both through device and lay-
out architecture improvements and through an optimization of the used PC
materials. As an example, in the field of embedded memories for automotive
systems the operating temperatures are close to the crystallization temperature
of Ge2Sb2Te5 (GST), so this material is unable to fulfill the requirements on
data retention and alternative materials must be used. Good PC materials can-
didates should have the properties listed in Table 1.1 and they must be able to
maintain these properties even when dimensions are reduced and the material
32
1.5 Goals and outline
gets more and more confined in the cell structure.
In the last years, possible alternatives to GST have been found in using
different compounds, as GeTe, or in doping GST with N. The fast transition
speed and good retention properties of GeTe makes it suitable for embedded
applications. Right before the beginning of this thesis work in has been shown
that C and N doping improves drastically the data retention of GeTe and lowers
the current needed for amorphization. However, up to now the effects of doping
on the structure of GeTe are still not clear, in particular on the stabilizing effect
on the amorphous phase. Moreover, the effect of confinement on PC materials
is still an open subject even for standard GST. It is also important to underline
that in a memory device the active layer is always in contact with different
interface materials that can eventually influence its properties, especially for
reduced dimensions.
For those reasons, the aim of this thesis is to study PC materials proper-
ties in three directions. First, the effect of C and N doping on the structure
of amorphous GeTe has been investigated and some interesting results on the
stability of the amorphous phase have been obtained. Second, the effect of three-
dimensional confinement has been studied on very small nanoclusters of GST,
deposited by sputtering with a new technique and characterized through X-Ray
diffraction. Third, the effect on the crystallization temperature of interfacing
GST and GeTe thin films with different materials has been studied through
reflectivity and X-ray diffraction measurements.
In Chapter 2 the study on the effect of doping on GeTe will be discussed.
From X-ray scattering experiments performed at the synchrotron SOLEIL (Saclay)
on amorphous powders of GeTe, C-doped GeTe and N-doped GeTe pair distri-
bution function (PDF) will be obtained. The PDF is proportional to the prob-
ability of finding two atoms at a certain distance. This quantity provides useful
information on the local structure of a material, which is the only exploitable
one for the amorphous phase. The experimental results will be interpreted and
discussed thanks to ab-initio simulations.
The effect of confinement on GST is the subject of Chapter 3. The de-
position of nano-sized clusters of GST with an average size of around 5.7 nm
33
1.5 Goals and outline
(± 1 nm) in a matrix of Al2O3 will be described. At the moment, those clus-
ters are the smallest GST clusters ever deposited by sputtering, and with the
narrowest size distribution. X-Ray diffraction measurement performed at the
ESRF synchrotron (Grenoble) in order to observe the crystallization of clusters
will be be reported and the discussions and conclusions will be focused on the
interpretation of the observed crystallization temperature.
In Chapter 4, the effect of interface layers on the crystallization of phase
change material will be investigated. This subject has not been largely treated
in literature up to know, so it is a new and very interesting field of research.
First, the measurements of the crystallization temperature of GST and GeTe
thin films embedded in three different materials (Ta, TiN and SiO2 ) through
reflectivity measurements will be reported. In the following, the study will be
focused on GeTe only and structure properties of the crystalline phase such as
the grain size and texture will be investigated through X-Ray diffraction analysis.
Some hypothesis will be presented on the nature of crystallization in presence
of different interfaces to support the conclusions on the obtained results.
34
Chapter 2
Effect of doping on the structure
of amorphous GeTe
35
2.1 State of the art on doping effects in phase change materials
2.1 State of the art on doping effects in phase
change materials
As already stated in the first Chapter, PCM must fulfill many requirements in
order to be competitive on the market. Those requirements include having a
good data retention and a low reset current, so that the data are preserved at
least 10 years at room temperature and the power required for programming
the cell in the RESET state is low. The data retention depends on the time
required for the metastable amorphous phase to crystallize, so it can be im-
proved by employing PC materials with a more stable amorphous phase and
a higher crystallization temperature Tx . The RESET current depends on the
resistivity of the crystalline phase because the reamorphization process requires
to heat the material by Joule effect through the programming current pulse, and
a more resistive material can be heated by lower current pulses. These goals can
be achieved by exploring new Ge-Sb-Te compound, but another possibility to
increase both the resistivity of the crystalline phase and the retention in the
amorphous state is opened by doping 1. The effect of introducing a dopant ele-
ment in PC materials has been widely investigated during the last two decades.
The most studied dopant is Nitrogen (N), but many other doping elements have
been studied lately, including Boron (B), Silicon Dioxide (SiO2 ) and, more re-
cently, Carbon (C).
The effect of N doping has been studied since late 90s, when it was observed
that cyclability for optical disks was improved by adding nitrogen into a Ge-
Sb-Te recording layer [34, 35]. When phase change electrical memories started
to be intensively studied, after year 2000, nitrogen-doped GST (NGST) gained
interest as a PCM device material due to its high crystallization temperature,
high retention time and high resistivity [36, 37]. NGST has been successfully
integrated into PCM devices [38, 39] and even in memory arrays [40]. The
literature about N-doped GST is quite vast but it will not be treated further
here, apart from a results on the structure of NGST reported in section 2.6.
1The term doping is conventionally used in this context to indicate the addition to the PC
material of elements in a concentration of several percents.
36
2.1 State of the art on doping effects in phase change materials
Figure 2.1: Low field cell resistance as a function of progam current for GST and GeTe cells
for various programming pulse times [24]. It can be noted that the SET operation for the
GeTe cell is faster and the difference in the resistance of the amorphous and crystalline phases
is higher.
As already stated in section 1.4, during the last years the GeTe compound
has been identified as a possible good candidate for embedded PCM due to its
higher crystallization temperature and better retention time than undoped GST
[22]. Moreover, GeTe PCM cells are characterized by a rapid SET operation
compared to GST cells (see Fig.2.1) and have demonstrated high cyclability
[24]. However, the required RESET current is the same for GST and GeTe.
In order to increase the resistivity of the crystalline cell and improve further
the properties of good retention and high Tx , N-doped GeTe has been studied.
N-doped GeTe (hereafter called GeTeN) with various percentages of N has been
successfully integrated into memory devices, with beneficial effects compared to
standard GeTe such as an even faster crystallization and a higher retention time.
The best performances were obtained for a N concentration of 2% [41], as shown
in Fig.2.2. An increase of the resistance of the crystalline phase for GeTeN
(N=8.4%) compared to undoped GeTe and an increase of the crystallization
temperature for GeTeN (N=9.81%) have also been observed [42, 43].
37
2.1 State of the art on doping effects in phase change materials
Figure 2.2: Calculation of the activation energy EA by interpolation of the fail times as a
function of 1/kT . In order to obtain the fail time, a PCM cell is written in the RESET state
and the fail time is defined as the time at which the resistance of the cell is reduced by one
half.
More recently, it has been shown that also C-doping can have a great impact
on the performances of GeTe. GeTeC devices with doping percentages of 4 and
10% showed an improved data retention in temperature compared to undoped
GeTe, indicating that the amorphous phase stability is improved by C-doping
[44]. As can be seen in Fig.2.3 and Table 2.1 [45], the crystallization temperature
of GeTeC is much higher than that of undoped GeTe and the effect of C-doping
is even stronger than the one of N-doping. For example, for GeTeC (C=10%)
Tx ≈ 325C , to be compared with Tx ≈ 275C of GeTeN (N=10%). Moreover,
the retention time, the activation energy EA and the crystal resistivity increase
with doping while the reset current decreases, as shown in Fig.2.4.
Doped phase change materials are thus promising candidates in order to fulfill
some requirements on PCM as the high data retention at elevated temperatures
or the low reset current. However, even if their good performances have been
demonstrated, a deeper understanding of the effects of doping on the structure
of GST and GeTe was lacking. Clarifying this subject can lead to a better
material engineering. The open questions, at least when this study has been
performed, include the impact of dopants on the amorphous structure and their
38
2.1 State of the art on doping effects in phase change materials
Figure 2.3: Reflectivity measurements of C and N doped GeTe films (150 nm thick) [45]. In
both cases Tx increases with increasing doping concentration and the effect is stronger for C
doping.
Figure 2.4: Activation energy (left) calculated for undoped and C-doped GeTe and low
electric field resistance as a function of the programming current (right) for a GST, undoped
GeTe and C-doped GeTe cell [44]. The activation energy increases and the RESET current
decreases with doping.
39
2.2 Theory of the Pair Distribution Function (PDF) g(r)
Material Tx [C ]
GeTe 175
GeTeC 5% 280
GeTeC 10% 320
GeTeC 15% 365
GeTeC 20% 380
GeTeN 5% 235
GeTeN 10% 265
GeTeN 15% 290
GeTeN 20% 310
Table 2.1: Crystallization temperatures Tx of C and N doped GeTe films (150 nm thick),
taken as the midpoint of the rising steps of the reflectivity curves reported in Fig.2.3 [45].
location in the crystalline structure. This is essential for understanding if the
structure of the materials remains stable after many cycles of crystallization and
reamorphization.
In the following, experimental results obtained through X-ray scattering on
the structure of amorphous C-doped and N-doped GeTe will be presented. Ab
initio simulations have also been performed in order to have a better under-
standing of the experimental results. The results presented in this chapter have
been published in Ref. [46].
2.2 Theory of the Pair Distribution Function
(PDF) g(r)
The amorphous phase is characterized by the absence of the long-range periodic
order that can be found in the crystal phase, but a short and medium-range order
is still present (see section 1.4.2). The determination of the Pair Distribution
Function (PDF) g(r) allows to have an insight on these local orders, describing
the correlations between pairs of atoms. In the following, the theory at the basis
of the PDF determination will be described [47, 48].
40
2.2 Theory of the Pair Distribution Function (PDF) g(r)
Figure 2.5: Schematic representation of an X-ray incident beam scattered by a point-like
sample. The incident wavevector is k0, the scattered wavevector is kf and the momentum
transfer is Q = k0 − kf .
Let us consider a X-ray incident beam on a point-like sample containing N
atoms as depicted in Fig.2.5. The incident beam is characterized by a wavevector
k0 of modulus 2π/λ and an energy E0. The scattered beam is characterized by
a wavevector kf and an energy Ef . The momentum transfer is Q = k0 − kf
and the energy transfer is hω = E0 −Ef . The double differential cross-section is
defined as the number of photons (or neutrons) scattered per unit of time into
the solid angle interval [Ω,Ω + dΩ] and into the energy interval [Ef ,Ef+dEf ]. For
X-Ray scattering, the incident beam energy is of the order of several keV so that
the maximum energy transfered between the incident photon and the sample is
negligible compared to the incident energy and k0 ≈ kf . As a consequence, the
modulus of Q is simply 4πsinθ/λ, with 2θ being the scattering angle, as shown in
Fig.2.5. In scattering experiments no analysis of the out-coming photon energy
is performed, so that all the photons in the solid angle dΩ are measured whatever
their energy, resulting in a cross-section
dσ
dΩ=
∫ E0
−∞
d2σ
dΩdEf
hdω (2.1)
The general expression of dσdΩ
is
dσ
dΩ=
⟨N∑
i,j
fi (Q) fj (Q) eiQ·rij
⟩
(2.2)
41
2.2 Theory of the Pair Distribution Function (PDF) g(r)
where fi (Q) is the atomic form factor of atom i at the position ri and rij = rj−ri.
In the equation, i can be equal to j. The notation 〈〉 in Eq. 2.2 denotes
an average on all possible positions of the scattering centers due to thermal
agitation. In order to understand the meaning of Eq. 2.2, let us first consider
the case of a monoatomic sample where all atoms have the same form factor
f (Q), so that
1
N
dσ
dΩ= f 2 (Q)
⟨
1
N
N∑
i,j
eiQ·rij
⟩
= f 2 (Q) · S (Q) (2.3)
where the so-called structure factor S (Q) is defined as
S (Q) =
⟨
1
N
N∑
i,j
eiQ·rij
⟩
(2.4)
where i can be equal to j. The structure factor S (Q) is a dimensionless quantity
and tends to 1 whenQ tends to∞. In a crystal, due to periodicity, S (Q) consists
of peaks for Q vectors belonging to the reciprocal lattice. In a disordered system
such as a liquid or a glass, the scattering is isotropic and the structure factor
only depends on the modulus of Q. It exhibits broad oscillations as it will be
seen on Fig.2.6. In the isotropic case, S (Q) can be written as
S (Q) =
⟨
1
N
N∑
i,j
sin (Qrij)
Qrij
⟩
= 1 +1
N
N∑
i 6=j
sin (Qrij)
Qrij(2.5)
where rij is the interatomic distance between atoms i and j. Considering that
also f (Q) depends only on the modulus of Q, Eq. 2.4 can be written as
1
N
dσ
dΩ= f 2 (Q) · S (Q) (2.6)
which can be rewritten as
1
N
dσ
dΩ= F (Q) + f 2 (Q) (2.7)
where F (Q) = f 2 (Q) [S (Q)− 1] is the so-called interference function. It is then
possible to describe the structure of the sample in the real space by means of the
pair distribution function g (r), which can be obtained by Fourier transformation
from the structure factor S (Q) as
g (r) = 1 +1
2π2rρ0
∫ ∞
0
Q [S (Q)− 1] sin (Qr) dQ (2.8)
42
2.2 Theory of the Pair Distribution Function (PDF) g(r)
where ρ0 =NVis the atomic number density of the sample. The pair distribution
function g (r) tends to zero for r → 0 and tends to 1 for r → ∞. It can be
demonstrated that
g (r) =1
N
⟨∑N
i 6=j δ (r − rij)⟩
4πr2ρ0(2.9)
In Eq. 2.9 the meaning of the function g (r) is clearly stated. It is proportional to
the probability of finding two atoms at a distance r. It should be also underlined
that 4πr2ρ0 · g (r) dr is the average number of atoms between r and r+ dr. It is
possible to obtain the structure factor S (Q) starting from g (r) by means of an
inverse Fourier transformation
S (Q)− 1 =4πρ0Q
∫ ∞
0
r [g (r)− 1] sin (Qr) dr (2.10)
In a polyatomic system constituted of n distinct chemical species denoted α, β,
etc. it is necessary to define partial pair distribution functions gαβ (r) and partial
structure factors Sαβ (Q). In the formalism developed by Faber and Ziman [49]
one defines
gαβ (r) =1
Ncαcβ
⟨∑
iα 6=jβδ(r − riαjβ
)⟩
4πr2ρ0(2.11)
Sαβ (Q) = 1 +1
Ncαcβ
⟨∑
iα 6=jβ
sin(Qriαjβ
)
Qriαjβ
⟩
(2.12)
where cα and cβ are the concentration of atoms for the species α and β so that∑n
α cα = 1, N is the total number of atoms, iα is an atom of species α, iβ is an
atom of species β and ρ0 =NV
as in the monoatomic case. The sum on iα runs
from 1 to Nα = cαN and the sum on jβ runs from 1 to Nβ = cβN . The partial
pair distribution function gαβ (r) is proportional to the probability of finding
an atom of species α at a distance r from one atom of species β. The partial
structure factor and the partial pair distribution function are related to each
other though the relations
Sαβ (Q) = 1 +4πρ0Q
∫ ∞
0
r [gαβ (r)− 1] sin (Qr) dr (2.13)
gαβ (r) = 1 +1
2π2rρ0
∫ ∞
0
Q [Sαβ (Q)− 1] sin (Qr) dQ (2.14)
43
2.2 Theory of the Pair Distribution Function (PDF) g(r)
It is possible to generalize Eq. 2.7 obtained in the monoatomic case as
1
N
dσ
dΩ=
∑
α,β
cαcβfα (Q) fβ (Q)Sαβ (Q)−∑
α
cαcβfα (Q) fβ (Q) +∑
α
cαf2α (Q)
=∑
α,β
cαcβfα (Q) fβ (Q)Sαβ (Q) +∑
α
cαf2α (Q)−
[∑
α
cαfα (Q)
]2
(2.15)
By defining⟨f 2⟩=
∑
α
cαf2α (Q)
and
〈f〉 =∑
α
cαfα (Q)
it is then possible to rewrite the cross-section in Eq. 2.15 as
1
N
dσ
dΩ= 〈f〉2 S (Q) +
⟨f 2⟩− 〈f〉2 (2.16)
where the total structure factor S (Q) is defined as the weighted sum of the
partial structure factors Sαβ (Q)
S (Q) =
∑
α,β cαcβfα (Q) fβ (Q)Sαβ (Q)
[∑
α cαfα (Q)]2(2.17)
It is then possible to define a total pair distribution function g (r) by Fourier
transformation in analogy with the monoatomic case (Eq. 2.8)
g (r) = 1 +1
2π2rρ0
∫ ∞
0
Q [S (Q)− 1] sin (Qr) dQ (2.18)
However, due to the Q dependence of the atomic form factors, g (r) has no direct
physical meaning. In particular it cannot be expressed as a weighted sum of the
partial pair distribution functions gαβ (r).
The total structure factor can be deduced from the measured scattered in-
tensity as a function of the scattering angle 2θ , after several experimental cor-
rections (see appendix A.2.2). It is important to underline that through the
experiment it is possible to measure only the total structure factor S (Q) and
not the partial ones. The atomic form factors increase with the electron number
in the atom (in the limit Q = 0 fα (0) = Zα, the number of electrons). From Eq.
44
2.3 Description of the samples
2.17 it can be deduced that, in presence of heavy elements, the contribution of
light elements to the total S (Q) will be relatively small, especially if they are
in low concentration. This will be the case in the following for all partial terms
involving C or N in carbon and nitrogen doped GeTe.
It is indeed possible to compute through ab initio simulations either all the
partial pair distribution functions gαβ (r) or the partial structure factors Sαβ (Q).
The latter can then be summed according to Eq. 2.17 to obtain a calculated
total structure factor Scalc (Q) which can be directly compared to the measured
one Sexp (Q). In order to compare experiments and calculations in r, the Fourier
transform of Scalc (Q) and Sexp (Q) must be calculated according to Eq. 2.18,
leading to gcalc (r) and gexp (r) respectively. However, gcalc (r) cannot be directly
obtained by summing calculated gαβ (r) as already remarked above.
2.3 Description of the samples
Amorphous thin films 200 nm thick of GeTe, C-doped GeTe containing 9.5 and
16.3 at. % of C and N-doped GeTe containing 4 and 10 at. % of N were deposited
by sputtering on a Si (100) substrate as described in appendix B. The Ge and Te
concentrations in the films were measured by Rutherford Back Scattering (RBS)
and the N and C dopant concentrations by Nuclear Reaction Analysis (NRA). In
all the studied samples, there is a small excess of Ge corresponding to Ge52Te48,
but in the following we will refer to the material as GeTe for simplicity. The
atomic number densities deduced from the measured mass densities of all the
samples are reported in table 2.2.
To perform the X-ray scattering experiments, the films have been gently
scratched in order to obtain a powder and the samples have been prepared by
putting about 1 mg of that powder inside a borosilicate glass capillary (700 µm
diameter). The packing fraction was of the order of 10%.
45
2.4 X-Ray scattering measurements and results
Material Mass density [g/cm3] Atomic number density [atoms/A3]
GeTe 5.38 0.0327
GeTeC9.5% 5.16 0.0342
GeTeC16.3% 5.12 0.0363
GeTeN4% 5.285 0.03328
GeTeN10% 5.22 0.0347
Table 2.2: Measured mass densities and atomic number densities for Ge52Te48, undoped and
doped with carbon or nitrogen, expressed in g/cm3 and atoms/A3, respectively. The mass
densities have been measured by X-ray reflectivity (XRR).
2.4 X-Ray scattering measurements and results
The X-Ray scattering experiments were performed on beamline CRISTAL at the
SOLEIL Synchrotron (Saclay, France) at an incident energy of E=45.4793 keV
(λ = 0.4441A) in transmission geometry. The transmitted scattered intensity
was collected by a 2D image plate detector. A photo and a schematic represen-
tation of the experimental setup are shown in appendix A.2.2 in Fig.A.4. The
capillaries filled with powder were installed on a small goniometer which was
rotating during the measurements in order to average possible non uniformity
of the powder in the capillary. For each sample the coefficient of transmission
was determined. The empty capillary signal has been measured to be used
for subtraction. Considering the distance D ≈ 21cm between the sample and
the detector, scattering by air is not negligible. Thus, the air signal has been
determined by measuring the scattered intensity without sample or capillary.
The intensity as a function of 2θ has been obtained for each sample by in-
tegrating the diffracting rings over a vertical sector of the 2D image and by
correcting the obtained intensity as explained in details in appendix A.2.2. The
resulting intensity as a function of 2θ can easily be transformed through the
relation Q = 4πsinθ/λ in a function of Q, I (Q), which is proportional to the
cross section 1N
dσdΩ. The total structure factor S (Q) can be obtained from I (Q)
through Eq. 2.16 after adequate normalization. The pair distribution function
g (r) is then calculated from S (Q) through Fourier transform (see Eq. 2.18)
46
2.4 X-Ray scattering measurements and results
over a range of Q, that in our case has been chosen between 0.1 and 12 A−1,
and by using the densities of Table 2.2.
The S(Q) and g(r) of the undoped GeTe sample are shown in Fig.2.6. They
both corresponds very well to those reported in Ref.[27]. From ab initio simula-
tions, [33] it is known that the first peak of g (r) (at about 2.6 A) is due to the
contributions of mostly Ge-Te bonds with only a few Ge-Ge bonds. The Te-Te
bonds give the principal contribution to the second peak of g (r) (around 4.1
A). In Fig. 2.7 the g(r) of amorphous GeTe is compared to that of crystalline
GeTe measured under the same experimental conditions. In crystalline GeTe,
the contribution to the first peak is due to the short and long Ge-Te distances
in the rhombohedral structure while the second peak is due to Ge-Ge and Te-Te
second neighbours (see Fig.1.11).
In Fig.2.8 the first two peaks of the pair distribution functions obtained for
the GeTeC (a) and GeTeN (b) samples are reported and compared to undoped
GeTe. In all cases, the first peak at around 2.6 A remains unchanged, indicating
that the first distances are not affected by doping, while there is an effect on
the second peak. A new peak appears at around 3.5 A in the doped samples
and the intensity of the peak at around 4.1 A is reduced. If one observes the
differences between the samples doped with different dopants concentrations it
is clearly visible that the effects increase as a function of doping and are more
marked for the GeTeC samples than for the GeTeN samples.
It is important to recall that through the experiment it is possible to measure
only the total functions and not the partial ones, as already underlined in section
2.2. The contribution of the various elements to the factors 〈f〉2 and 〈f 2〉 of
Eq. 2.15 is proportional to the concentration of each element and to its atomic
number Z. Thus, the contributions of bonds involving C or N is negligible due
to the low atomic number and weak concentration of C and N atoms compared
to Ge and Te atoms. For instance, in C-doped GeTe containing 16.3 at. % of
C the contribution of pairs involving C (Ge-C, Te-C and C-C) to the measured
g(r) is much smaller (respectively, 1.6%, 2.8% and 0.04%) than that of Ge-
Ge (12.5%), Ge-Te (44.1%) and Te-Te (38.8%) pairs. Only the Ge-Ge, Ge-Te
and Te-Te bonds contribute to the total pair distribution function determined
47
2.4 X-Ray scattering measurements and results
0 2 4 6 8 10 12 14 16 18 200.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
S(Q
)
Q[Å-1]
2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
(b)
g(r
)
r[Å]
(a)
Figure 2.6: Measured (a) S (Q) and (b) g (r) for undoped amorphous GeTe. It can be noted
that S (Q) tends to 1 for high Q.
48
2.5 Ab initio simulations
2.0 2.5 3.0 3.5 4.0 4.5 5.0-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
g(r
)
r[Å]
Figure 2.7: Comparison between the measured g(r) of amorphous (blue) and crystalline
(red) undoped GeTe.
experimentally and one can conclude that the new peak appearing for the doped
samples is not due to bonds directly involving C or N.
2.5 Ab initio simulations
Considering that only the total pair distribution functions can be measured ex-
perimentally, it is impossible to deduce the origin of the peak of g(r) that appears
for doped samples. In order to go further in the analysis, ab initio molecular
dynamics simulations were performed for both GeTeC and GeTeN by Dr J-Y.
Raty of the University of Liege. The simulations were done on 210 atom boxes
(92 Ge, 86 Te, 32 C for the C-doped sample and 97 Ge, 92 Te, 21 N for the
N-doped sample) corresponding to Ge52Te48 doped with 15 at. % C and 10 at.
% N. The initial positions of Ge and Te atoms were those found in a previous
simulation of amorphous GeTe [50]. Then C (or N) atoms were introduced by
randomly substituting Ge and Te atoms. The electronic properties were com-
49
2.5 Ab initio simulations
2.0 2.5 3.0 3.5 4.0 4.5 5.0-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
g(r) GeTe
g(r) GeTeC 9.6%
g(r) GeTeC 16.3%
g
(r)
r[Å]
2.0 2.5 3.0 3.5 4.0 4.5 5.0-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
g(r) GeTe
g(r) GeTeN 4%
g(r) GeTeN 10%
(b)
g(r
)
r[Å]
(a)
Figure 2.8: Measured g (r) for (a) undoped GeTe and GeTeC (C=9.6% and 16.3%) and (b)
undoped GeTe and GeTeN (N=4% and 10%). In both cases, the first peak is constant with
doping while the intensity of the second peak of the undoped sample decreases with increasing
doping contents. A new peak appears at around 3.5 A in the doped samples. These effects
increase as a function of doping and are stronger in the GeTeN case.
50
2.5 Ab initio simulations
puted within density functional theory (DFT) implemented in the VASP code,
using PAW potentials and the PBE-GGA exchange correlation as described in
Ref.[51]. The liquid system was simulated at 3000 K for 3 ps, and then ther-
malized at 1073 K for 10 ps. It was cooled down to the final temperature of
300 K with a 30 K/ps ramp. The density was adjusted to minimize the effective
stress on the simulation box. The refined atomic densities are 0.033 atoms/A3
for undoped GeTe, 0.0343 atoms/A3 for GeTeC15% and 0.0337 atoms/A3 for
GeTeN10%, close to what measured experimentally. The amorphous structural
data were gathered by averaging relevant quantities over a 10 ps trajectory at
300 K.
From the simulation it is possible to obtain all the partial pair distribution
functions gαβ (r) and so it is possible to evaluate the contribution of each atomic
pair, including the ones involving light elements that could not be observed
experimentally. The total stucture factors Scalc (Q) have been obtained from
weighted partial structure factors as described in Eq. 2.17. The pair distribu-
tion functions gcalc (r) have been then calculated from Scalc (Q) though Fourier
transform as described in Eq. 2.18, using the same window 0.1< Q <12A as
for the experimental data, and the results are shown in Fig.2.9. For undoped
GeTe, the calculated gcalc (r) is in good agreement with literature [28, 33, 52].
In the calculated gcalc (r) the first peak remains almost constant with doping,
the second peak decreases with doping and a new peak appears around 3.3A for
the doped samples. The effects of C and N doping on the amorphous structure
of GeTe are comparable.
By comparing the simulated and experimental g (r) in Fig.2.9, a shift in the
peak positions can be observed. The existence of this shift is already known
from literature [53, 54]. Besides, in the experimental g (r) the effect of N doping
is weaker than the one of C doping. However, apart from those two differences,
the good agreement between the evolution as a function of doping for the sim-
ulated and measured total pair distribution functions allows to conclude that
the ab initio simulation takes properly into account the effect of dopants on the
amorphous structure of GeTe.
The next step is to analyze the information that can be obtained from a visual
51
2.5 Ab initio simulations
2.0 2.5 3.0 3.5 4.0 4.5 5.0
0
1
2
3
4
5
6
7
8
g(r) GeTe
g(r) GeTeC 16.3%
g(r) GeTeN 10%
Simulation
g(r
)
r[Å]
Experiment
Figure 2.9: Comparison between measured and calculated g (r) for undoped GeTe, GeTeC
(C=16.3% in the experiment and C=15% in the simulation) and GeTeN (N=10% both in the
experiment and in the simulation). Even if an already known shift between peaks positions
can be observed, the evolution of the simulated and measured pair distribution functions with
doping are in good agreement. The effect of N-doping is stronger in the calculated g (r) than
in the measured one.
52
2.5 Ab initio simulations
1 2 3 4 5
0
1
2
3
4
5
6
7
8
9
10
11
12
13
1 2 3 4 5
0
5
10
15
20
25
30
35
40
45
50
(b)
(a) GeTe
GeTeC 15%
GeTeN 10%
Ge-Ge
Te-Te
g(r
)
r[Å]
Ge-Te
C-Te/N-Te
C-Ge/N-Ge
C-C/N-N
g(r
)
r[Å]
GeTeC 15%
GeTeN 10%
Figure 2.10: Partial pair distribution functions for (a) Ge-Ge, Te-Te, and Ge-Te pairs in
doped and undoped samples and (b) pairs involving C or N. Curves are shifted for clarity. A
new peak appears in the range 3.1-3.5 A in the Ge-Ge partial pair distribution function for
both C doped and N doped samples, while it is absent in the undoped sample. A difference
between partial contributions involving C and the ones invloving N is the absence of Te-N
bonds at small distances (less than 3.2 A).
53
2.5 Ab initio simulations
inspection of the simulation boxes and from the calculated partial pair distribu-
tion functions in order to understand the local atomic arrangements. The partial
pair distribution functions calculated for GeTe, GeTeC15% and GeTeN10% are
shown in Fig.2.10. For simplicity, in Fig.2.10(a) the pair contributions due to
Ge-Ge, Ge-Te, and Te-Te pairs in doped and undoped samples are reported,
shifted for clarity, while the pairs involving C or N are reported in Fig.2.10(b).
From Fig.2.10(a) it is evident that the contributions to the first peak are given by
a few Ge-Ge bonds, at an average distance of 2.63 A for GeTe and GeTeC sam-
ples and 2.65 A for the GeTeN sample, and mostly by Ge-Te bonds at an average
distance of 2.74 A in GeTeC and 2.76 A in GeTe and GeTeN. For both doped
and undoped GeTe the most probable Te-Te distance is rather large (around
4.1 and 4.2 A), with only a few Te-Te bonds about 2.95 A. This corresponds
to a tendency for a strong chemical order due to the alternation of Ge and Te
atoms that is the precursor of the rhombohedral crystalline GeTe structure [55]
and this tendency is maintained in presence of doping. The most relevant result
is the appearance of a new peak at around 3.39 A in the Ge-Ge partial pair
distribution function for both C doped and N doped samples, while it is absent
in the undoped sample.
Let us focus first on the analysis of the C-doped sample. From Fig.2.10(b) the
presence of C-C bonds at a very short distance (1.31 A) can be noticed. These
bonds were not present in the initial state of the system and correspond to a
p-type of bonding. By observing the snapshot of the final state of the simulation
box for GeTeC reported in Fig.2.11, C chains of various lengths (2−5 atoms) can
be observed, without any branching. Observing Fig.2.10(b) it is also possible to
notice many C-Ge bonds that contribute to the partial pair distribution function
at an average distance of 2.05 A. Also a few C-Te bonds are present, with a
contribution at around 2.23 A, but they are much less frequent than C-Ge bonds.
The analysis of bond angle distribution shows that Ge atoms for undoped
GeTe are either in a tetrahedral environment (about 25%-35%) or in a strongly
distorted octahedral environment. In C-doped GeTe, the bond angle distribu-
tion becomes broader making the distinction between tetrahedral and octahedral
environment less clear. The inspection of the model structure (Fig.2.11) com-
54
2.5 Ab initio simulations
Figure 2.11: Snapshot of the final state of the simulation box for GeTeC. Ge atoms are
represented in pink, Te atoms in light blue and C atoms in red. The inspection of this box
combined with a bond angle analysis around C atoms, reveals the presence of a mixture of
tetrahedral (C−TeGe3, C−Ge4 and C−Ge2Te2), triangular (C−C−Ge2 and C−C−GeTe),
and linear (C chains) bonds.
55
2.5 Ab initio simulations
Figure 2.12: Summary of the carbon environments in the C-doped GeTe sample. C−TeGe3,
C−Ge4 and C−Ge2Te2 tetrahedra can be found, as well as C−C−Ge2 and C−C−GeTe
triangular environments.
bined with a bond angle analysis around C atoms reveals the presence of a
mixture of tetrahedral (sp3), triangular (sp2), and linear (sp) C-Ge bonds. The
C−TeGe3 tetrahedra are the majority (3 are present) and one C−Ge4 and one
C − Ge2Te2 tetrahedra can also be found, in agreement with the relative ratio
of C-Ge and C-Te bonds. Eleven C−C−Ge2 and one C−C−GeTe triangular
environments are also present. A schematic representation of the environments
involving carbon is depicted in Fig.2.12. Two Ge atoms belonging to the same
tetrahedral unit centered on a carbon atom result to be separated on average
by 3.39 A and that is the same distance for two Ge of a C − C − Ge2 triangle.
This Ge-Ge distance is induced by the presence of C and it can be clearly seen
in Fig.2.10(a), where a new peak appears at 3.39 A in the Ge-Ge partial pair
distribution of GeTeC. In the r-range from 3.2 to 3.6 A the Ge-Te and Te-Te
partial pair distribution are only slightly affected by C doping.
Let us now analyze the N-doped sample. By comparing with GeTeC, the
first striking difference that can be noticed is the absence of Te-N bonds at
small distances (less than 3.2 A) for GeTeN while there are a few Te-C bonds
56
2.5 Ab initio simulations
Figure 2.13: Snapshot of the final state of the simulation box for GeTeN. Ge atoms are repre-
sented in pink, Te atoms in light blue and N atoms in green. N−Ge3 pyramidal environments,
N−Ge4 tetrahedral environments and N2 molecules can be found.
57
2.5 Ab initio simulations
Figure 2.14: Summary of the nitrogen environments found in the N-doped GeTe sample.
N−Ge4 tetrahedra, N−Ge3 pyramids and N2 molecules have been observed.
at around 2.23 A in GeTeC. The only bonds involving N that can be found
in GeTeN for r < 3.2A are Ge-N and N-N bonds. In the snapshot of the
final state of the simulation box for GeTeN reported in Fig.2.13, 8 N − Ge3
pyramidal environments, 9 N − Ge4 tetrahedral environments and 2 molecules
of N2 can be found. The observed environments for N-doped GeTe are reported
in Fig.2.14. It should be noticed that during the simulated annealing procedure,
the initial configuration was taken from a liquid sample equilibrated at a very
high temperature (3000 K), so that any kind of possible bond was present in the
structure, including Te-N bonds. However, due to the low N concentration, no
N2 molecule was present in the liquid. It is thus striking to find that all the Te-N
bonds have been disrupted upon cooling and statistically unlikely N2 molecules
have formed.
58
2.6 Discussion
2.6 Discussion
The two main conclusions of this work are that, on one hand, the strong chemical
order between Ge and Te characteristic of amorphous GeTe is maintained in
presence of C or N doping elements and that, on the other hand, dopants deeply
modify the structure of the amorphous phase by introducing new environments.
A common feature of the C and N doping is the appearance of tetrahedral
units centered on C or N. In C-doped amorphous GeTe, triangular environments
around C and short C chains (containing between 2 and 5 C atoms) can also be
found, while in N-doped GeTe NGe3 pyramids and N2 molecules are also present.
One major difference between N and C doping is the fact that C can form short
bonds with Te, although in smaller proportion than C-Ge and C-C bonds, while
short N-Te bonds are absent in N-doped GeTe. The same conclusion has been
recently obtained in Ref. [56] through X-ray absorption spectroscopy and X-ray
photoemission spectroscopy (XPS). The fact that C is bonded preferably to Ge
has also been found by ab initio simulations in the case of C-doped GST in
studies performed at the same time as the study reported in this thesis [57, 58].
In N-doped GST, ab initio simulations show that N is bonded in majority to
Ge, to Sb in a less extent and marginally to Te [59].
In C-doped GeTe, two Ge atoms belonging to the same tetrahedral or tri-
angular unit are separated by 3.4 A. This Ge-Ge distance, that did not exist
in undoped amorphous GeTe, appears as a new peak in the Ge-Ge partial pair
distribution function and is also visible in the calculated total g (r). From these
ab initio simulations, it can be concluded that the additional peak observed
around 3.3 A in the measured pair distribution function of GeTeC (C=9.5%)
and GeTeC (C=16.3%) is mainly due to this new Ge-Ge distance present in
tetrahedral and triangular units centered on carbon. The contribution of Ge-Te
and Te-Te distances to the total g (r) in this r-range is much less than that of
Ge-Ge distances because of the small number of C-Te bonds compared to C-Ge
bonds. In N-doped GeTe two Ge atoms belonging to the same tetrahedral or
pyramidal unit are separated by 3.25 A. They generate a new peak in the Ge-Ge
partial pair distribution function and in the calculated total g (r).The effect on
59
2.6 Discussion
the measured g (r) of GeTeN (N=10%) is smaller than predicted by the calcu-
lation. One explanation could be that the sample contains more N2 molecules
than the simulation box so that the relative abundance of NGe4 and NGe3 units
is reduced in the sample. Information from literature on the presence of N2
molecules in N-doped amorphous GeTe is contradictory at the present time. N2
molecules have been found by X-ray absorption and XPS experiments in Ref.[60]
but not in [56].
It is worth noting that the samples studied experimentally in this work are
as-deposited amorphous samples, while the simulation results are collected on
an amorphous state obtained by a rapid quench of the liquid phase (the so
called melt-quenched amorphous). The structure of these two amorphous states
could differ. This question has been addressed in literature in the case of N-
doped GeTe by ab initio simulations [54]. The conclusion of this study is that
the difference in the local structure between as deposited and melt-quenched
amorphous phase is only the number of tetrahedral environments.
One can wonder whether the structural changes observed in doped GeTe can
be related to the enhanced crystallization temperature and activation energy of
C and N-doped GeTe. The presence of C or N makes the Ge environments, on
average, slightly more tetrahedral. According to the ideas developed in Ref.[50],
this structural change would increase the average number of constraints so dop-
ing could indeed increase the stability of the amorphous state. This idea led
to a study of the vibrational properties of C and N-doped amorphous GeTe
by combining Fourier Transform Infrared spectroscopy (FTIR) experiments and
computation of the vibration modes from the ab initio simulations presented in
section 2.5. The main conclusion of this study (Ref.[61], submitted for publica-
tion) is that the inclusion of C and N creates high frequency, localized modes,
and at the same time decreases the density of low frequency acoustic-like modes.
In Ref.[61] it is shown that this kind of effects corresponds to an increased sta-
bility of doped amorphous GeTe. Besides, in the case of C-doped GeTe the
presence of C chains could be an obstacle for the crystallization of the GeTe
matrix, provided that these C-C bonds disappear in the crystalline phase, since
breaking a C-C bond is an extremely endothermic process.
60
2.7 Conclusions and perspectives
2.7 Conclusions and perspectives
In conclusion, the addition of carbon or nitrogen in GeTe has been observed to
deeply modify the structure of the amorphous phase. Both dopants induce the
formation of new environments and bonds, resulting in a more stable amorphous
structure , which could explain the increase of the crystallization temperature
and activation energy induced by doping. In parallel to the work described in
this chapter, the effect of carbon doping on the properties of GST has been
observed experimentally very recently [62]. In that case ab initio simulations
[57, 58] have preceded the experimental study. The conclusion of all these works
is that carbon doping enhances the stability of the amorphous GST phase, like
in the case of C-doped amorphous GeTe studied in this thesis.
An enhanced stability of the amorphous phase would produce an increase of
the retention time, beneficial for PCM. Such an increase is indeed observed in
PCM containing C on N doped GeTe [44, 41]. However, understanding its origin
requires further studies since other effects could contribute as well. From the
device point of view, the capability of the material to maintain its properties after
many phase transformations is an imperative requirement. Therefore, starting
from the understanding of the amorphous structure of C and N doped amorphous
GeTe, the next step must be the study of the crystalline phase. At the moment it
is unclear if the doping elements remain in the crystalline phase, at substitution
or interstitial sites, or if they segregate, for instance at the grain boundaries. In
the latter case one can wonder if they are incorporated again in the amorphous
phase upon reamorphization.
Little is known on the crystalline structure of C and N doped GeTe. One
result is that the crystalline phase is cubic at room temperature for C and N
content above 10% [60, 63]. In the case of crystalline N-doped GST (NGST) dis-
tortions of the crystal lattice and reduction of the grain size have been observed
[34, 64]. The conclusion was that the N atoms occupy tetrahedral interstitial
sites in the NGST crystalline structure and that a Ge3N4 phase is formed near
the grain boundaries, thus reducing the grain size. Moreover, the presence of
some N2 molecules was also observed. The molecular nitrogen is believed to
61
2.7 Conclusions and perspectives
exist at interstitial and vacancy sites, and more likely at grain boundaries [37].
Further studies are required in order to localize the dopants in the crystalline
phase. Characterization techniques such as X-ray photoemission spectroscopy
and Fourier transform infrared spectroscopy will be useful to provide a detailed
comparison of the amorphous and crystalline phases and to investigate the evolu-
tion between them. Finally, local analysis via transmission electron microscopy
could be useful to study amorphous and crystalline phase in both thin films and
integrated materials.
62
Chapter 3
Confinement of phase change
materials: Ge2Sb2Te5 nanoclusters
63
3.1 Introduction
3.1 Introduction
One of the open questions about the competitiveness of PCM is at which extent
their dimensions can be reduced. Their theoretically very high scalability is one
of their most promising properties, as it can make them become competitive with
Flash memories. In terms of devices, the scaling of the memory cell is limited by
architectural and geometrical issues and by material scaling possibilities. Thus,
it is important to investigate the effect of shrinking size on the phase change
materials properties and characteristics. First, it is of the uttermost importance
to establish if there is a limit size under which the phase transformation can no
longer occur. Moreover, as it will be discussed further in the following sections,
the effect of reduced dimensions on the mechanisms of crystallization is still an
open question. This confinement effect can be studied at different extents: the
first approach is to investigate the effect of reducing thickness of PC thin films
on the phase change mechanism. A further step consists in investigating con-
fined nanostructures of PC materials, such as nanowires and nanoclusters. In
this chapter, an overview of the results obtained on confined structures will be
presented. A new method for the deposition of PC nanoparticles of diameter
below 10 nm will be introduced and results obtained on Ge2Sb2Te5 (GST) parti-
cles embedded in Al2O3 will be presented. The results presented in this chapter
have been published in Ref. [65].
3.1.1 Effect of shrinking size in one dimension: thin films
Many studies regarding the scalability of PCM deal with the thickness depen-
dence of the amorphous to crystalline phase transition in PC thin films, meaning
the effect of confining the material in only one dimension. In the following, a
review of the main results from literature on GST will be presented. First, evi-
dences of the influence of thin films thickness on the crystallization temperature
Tx will be shown, then the crystallization mechanisms will be discussed and at
the end some interesting results on multilayers will be reported.
In literature, the crystallization behavior of a PC material has been inves-
tigated by several experimental methods, as for example optical measurements
64
3.1 Introduction
[66], electrical measurements [67], Transmission Electron Microscopy (TEM)
[68], Differential Scanning Calorimetry (DSC) [69] or X-Ray Diffraction (XRD)
[70]. All the quantities measured by these methods are directly proportional to
the quantity of crystalline matter in the sample, except the case of resistivity
measurements where the value of resistivity decreases only when a crystalline
path between electrodes is formed. There is no generally admitted definition of
the crystallization temperature Tx and different criteria are used. For example,
it can be defined as the temperature at which the first derivative of a reflectivity
curve with respect to temperature has a maximum value, or as the point at
which the resistance of a sample is reduced by a factor of 2, or as the peak point
in a Differential Scanning Calorimetry experiment. The experimental conditions
are different for all these techniques, and considering that crystallization is ruled
by kinetic theories, the value of Tx depends on the thermal history of the sample
and thus on these conditions. This means that the value of Tx depends on both
the experimental conditions and on the criteria used to define it.That said, under
the same experimental conditions, it is widely assumed that a PC material film
at least 100 nm thick should be characterized by a crystallization temperature
Tx that is independent form the surrounding layers and can be considered as its
bulk crystallization temperature.
In 2007, Wei et al. measured the amorphous to fcc rocksalt phase transition
temperature, the crystallization speed and activation energy (Kissinger method)
for thin films of GST of various thicknesses (5, 10, 15, 20 and 30 nm) sandwiched
by 50 nm of ZnS − SiO2 [67]. The crystallization temperature Tx increases
as the film thickness d is reduced below 20 nm, as can be seen in Table 3.1,
while the crystallization speed decreases (as shown in Figure 3.1). No thickness
dependence of parameters has been found for films thicker than 20 nm. The
estimated crystallization activation energy increases with decreasing thickness,
from 2.86 eV for 20 nm thick films to 4.66 eV for 5 nm thick films, and a linear
relationship has been found between ln[Tx] and the thickness d of the GST thin
films. Moreover, the Avrami coefficient n (described in section 1.3.5) has been
measured and it has been found that n > 1.5 for d ≥ 20 nm, while 1 < n < 1.5
for d = 10 nm and n < 1 for d = 5 nm. This suggests that the nucleation and
65
3.1 Introduction
Figure 3.1: Resistivity as a function of time at room temperature for thin films of GST of
different thicknesses pre-annealed at 143.5C (from Ref.[67]). The incubation time τ , defined
as the time elapsed before the onset of crystallization, and the transition time from the highest
to lowest resistivity increase with decreasing film thickness, meaning that the crystallization
speed is reduced for small thicknesses.
Figure 3.2: Model used Ref.[71] to interpret the thickness-dependent variation of Tx . A
cylindric crystalline nucleus is embedded in the amorphous phase, sandwiched between two
oxide interfaces.
66
3.1 Introduction
Heating rate (C /min) 5 nm 10 nm 15 nm 20 nm 30 nm
0.5 157 149 141 138 138
1 159 153 147 142 142
3 164 157 152 148 148
10 166 160 155 154 154
20 170 164 160 157 157
Table 3.1: Crystallization temperatures as a function of heating rate and GST film thickness
from Ref.[67].
growth mechanisms change by decreasing thickness.
In order to explain the thickness-dependent increase of Tx for thin films of Si
with oxide interfaces a model has been proposed by Zacharias [71] and it has been
be applied to PC materials. In this model, based on homogeneous nucleation,
an effective interfacial energy between the oxide capping layer and the surface
of a growing crystalline cylindrically shaped nucleus in the amorphous matrix
(see Fig.3.2) is introduced. This effective interfacial energy is a function of the
distance between the nucleus surface and the oxide. By calculating the Gibbs
free energy for the nucleus, the crystallization temperature Tx can be expressed
as an exponential function of the phase change film thickness d
Tx = Tac
[
1 +
(γoc − γac − γoa
γace−d/4l0
)]
(3.1)
where γac, γoc, and γoa are defined as the interfacial free energies per unit area
between the amorphous and crystalline phases (ac), between the oxide capping
layer and the crystalline phase (oc), and between the oxide capping layer and the
amorphous phase (oa), respectively. Tac is the bulk crystallization temperature
and l0 can be interpreted as an average screening or bonding length which is
related to the range of interatomic forces typical for the materials o and c. The
values reported in Table 3.1 can be fitted by Eq. 3.1, as reported in Ref.[67].
Another study on the effect of reducing thin film thickness over the crystal-
lization temperature Tx has been published by Raoux and coworkers [70]. The
value of Tx has been measured for films of various thicknesses (from 1 to 50
nm) and different materials (GST, N-doped GST, Ge15Sb85, Sb2Te and Ag- and
67
3.1 Introduction
In-doped Sb2Te). All the samples were deposited on Si and capped with 10
nm of Al2O3 . They have been characterized through X-Ray diffraction during
in-situ annealing with a heating rate of 1C /s. The transition temperature Tx ,
corresponding to the amorphous to fcc rocksalt structure transformation, is sim-
ilar to that of the bulk (close to 155C ) for all films with thicknesses above 10
nm. Tx increases with decreasing thickness below 10 nm for all the considered
materials, but at a different extent for each composition. The value reported
for Tx is 342C for a 5 nm thin GST film and 380C for a 2 nm thin GST film,
where the phase transition proceeds directly from the amorphous to the hexag-
onal phase. It can be noted that the variation of Tx with thickness for the GST
films capped with Al2O3 of Ref. [70] is much higher than for the GST films of
Ref. [67] capped with ZnS−SiO2. It is also worth noting that the so called bulk
crystallization temperature is 155C in Ref.[70] and 142C in Ref.[67] for the
same heating rate. This could come from the different experimental methods
(resistance measurements in Ref.[67] and X-ray diffraction in Ref.[70]). Other
possibilities are that the composition of the material is not exactly the same
(there can also be an inclusion of undesired doping elements) or that the dif-
ferent interface materials still influence Tx in films of 30-50 nm thickness (see
Chapter 4). The thickness dependence of Tx in Ref. [70] have been interpolated
using Eq. 3.1 [71] as it is shown in Fig.3.3. On the other side, the transition
from the cubic phase to the hexagonal phase occurs at the same temperature for
all film thicknesses, around 450C . The thinnest GST film that shows evidence
of Bragg peaks is 2 nm thick, so the authors of Ref. [70] conclude that this
thickness is the size limit for the existence of a phase change transformation for
GST.
In 2009, Simpson et al. measured the crystallization temperature of thin
films of GST of thicknesses between 2 nm and 10 nm encapsulated by TiN or
ZnS− SiO2 [72]. In the case of TiN an increase of Tx with decreasing thickness
has been obtained, similar to the trend observed in Ref.[70] for GST encapsu-
lated in Al2O3 , while for encapsulation with ZnS− SiO2 Tx is decreasing with
thickness. The latter result is in contradiction with what observed in Ref. [67]
and reported in Table 3.1. Both in Refs. [72] and [67] the variation of Tx with
68
3.1 Introduction
Figure 3.3: Crystallization temperature Tx as a function of film thickness for various PC
materials: GST, N-doped GST (NGST), Ge15Sb85 (GeSb), Sb2Te and Ag- and In-doped Sb2Te
(AIST) deposited on Si and capped with Al2O3 , fitted using Eq.3.1, as presented in Ref.[70].
69
3.1 Introduction
decreasing thickness for GST interfaced with ZnS − SiO2 is small, but in Ref.
[72] Tx increases with decreasing thickness while the opposite trend is observed
in Ref. [67]. The authors of Ref. [72] proposed the existence of a correlation be-
tween the mechanical stresses induced by the cladding material and the change
in Tx . One difficulty with this interpretation is that, even in the case where
the stress in the cladding layer is known, the evaluation of the level of strain
induced in the embedded layer is delicate [73]. Other recent studies indicate that
compressive and tensile stress inhibit the rocksalt to hexagonal phase transition
when the GST film thickness is thinner than 10 nm [74].
From all these results it is evident that the effect of confinement on phase
change cannot be separated from the effect of interaction between the interface
material and the PC material itself. The influence of interface materials on crys-
tallization of an amorphous compound had been studied also out of the PCM
context. For example, in 1969 Oki et al. reported that a 10− 30 nm film of Ge
crystallizes at different temperatures if interfaced with a metal [75]. In a study
on Pb/Ge multilayer samples in 1987 [76] a decrease of Tx with decreasing Ge
thin films thickness has been observed, and it has been suggested that the crys-
tallization is interfacially initiated and strongly affected by the Pb layer texture.
In late 1990s, the effect of the interface layer on crystallization mechanism of PC
materials has been extensively studied by Ohshima [66, 77]. The aim of those
studies was to understand the effect of the interface dielectrics commonly used
as capping layers for optical devices due to their optical good properties. Those
kinds of dielectrics include SiO2, Si3N4, Ta2O5, ZnS and ZnS − SiO2. The PC
materials are thin films of 30 nm with variable composition in the Sb2Te3−GeTe
pseudobinary system (named Ge−Sb−Te materials). The results indicate that
the crystallization temperature, the crystallization activation energy and the
nucleation process are affected by the different dielectric interfaces. The effect
of the dielectric films is to accelerate the nucleation in Si3N4 and Ta2O5 capped
samples, to inhibit the nucleation in SiO2 capped sample, and to generate nuclei
even during the grain growth process in ZnS and ZnS − SiO2 capped samples.
The author suggests that these variations may depend on the surface reactiv-
ity and chemical affinity of the film materials [66]. This conclusion has been
70
3.1 Introduction
supported by a further study on Ge − Sb − Te film interfaced with Si3N4 or
ZnS − SiO2, where a difference in the crystalline structure and in the grain
growth process has been evidenced between the two different samples [77]. The
films interfaced with Si3N4 crystallize in a fcc structure and the grain grow grad-
ually until the fcc to hexagonal transition, while for the films interfaced with
ZnS− SiO2 the structure was a mixture of fcc and hexagonal structure and the
grains grew abruptly at around 250C .
In 2009, Jang et al. investigated GST/SiO2 multilayered films with layer
thicknesses of [10.42 nm/10.42 nm]×5 (named M10) and [5.93 nm/5.64 nm]×10
(named M5) [78]. Sheet resistance measurements show that the phase-change
characteristics are affected by the bilayer thickness, as can be seen in Fig.3.4.
In particular, the crystallization temperature seems to be slightly higher for
multilayer films than for a single-layered GST thin film of 100 nm. Compared to
the single-layered GST thin film, the temperature region between the amorphous
to fcc transition and the fcc to hexagonal transition is reduced and it decreases
further as the bilayer thickness of the multilayered structure decreases. A similar
reduction was observed in Ref.[70] for GST with Al2O3 capping, but in that case
the amorphous to fcc transition temperature increased while the fcc to hexagonal
transition temperature remained unchanged, while in Ref.[78] the amorphous
to fcc transition temperature increases only slightly and the fcc to hexagonal
transition occurs at lower temperatures. The difference between the two studies
is again the capping material, that can have a different influence on the two
transitions. Moreover, in Ref.[78] it is reported (without quantitative data) that
the GST film is highly strained for thin multilayers due to the different thermal
expansion coefficients of GST and SiO2.
In conclusion, results from literature on the variation of Tx with thickness
are often unclear and ambiguous, and sometimes even contradictory. The inter-
face materials play certainly an important role in determining the amorphous to
fcc and the fcc to hexagonal transition temperatures, and can maybe have dif-
ferent influences over the two of them. Another important issue to be taken into
account is the possibility of interdiffusion from the capping layer through the PC
material, resulting in a doping effect instead of an interface effect. Size reduction
71
3.1 Introduction
Figure 3.4: Sheet resistance of multilayered films of GST/SiO2 as a function of annealing
temperature, as reported in Ref.[78]. The label M25, M10 and M5 indicate different bilayer
thicknesses (M5 corresponds to the thinnest sample, M25 the thickest one). The dotted lines
correspond to ex situ annealing temperatures used for further analysis in Ref.[78].
effects have been reported for thin films but while in some cases the variation of
Tx is extremely high, in other cases it is very small and not even consistent in
different publications (as for the ZnS−SiO2 interface in Ref.[72] and [67], as dis-
cussed above). The greatest variations of Tx have been reported for Al2O3 [70]
and TiN [72] interfaces, but there are no further studies in literature for those
materials. Some attempts to identify a relation between surface reactivity and
chemical affinity of films and a change in the crystallization mechanisms with
different interface materials have been done [66]. However, the physical and
chemical influence of different capping layers on the phase change mechanism it
is still unknown.
3.1.2 Effect of shrinking size in two and three dimen-
sions: nanostructures
As it was discussed in the previous section, reducing the thickness of PC material
thin films can have a great impact on crystallization mechanism. This has been
observed through the variation of some important phase change properties such
as crystallization temperature or activation energy, which depend also strongly
on the interface materials. However, the studies reported in the previous section
deal only with the effect of confining the PC materials in one dimension. To go
72
3.1 Introduction
Figure 3.5: Scanning Electron Microscopy (SEM) image of as-grown GST nanowires from
Ref.[81].
further, systems of nanostructures can be investigated. For example, nanowires
with length much higher than their diameter can be used to study the effect of
confinement in two dimensions.
In 2006, GeTe and Sb2Te3 nanowires have been obtained through vapor-
liquid-solid (VLS) technique [79, 80]. A Scanning Electron Microscopy (SEM)
image of as-deposited GST nanowires taken from Ref.[81] is reported in Figure
3.5. Interesting results about scaling effects on physical parameters have been
obtained with GST nanowires of different thicknesses, from 20 to 190 nm, fabri-
cated through VLS process [82] on a SiO2 /Si substrate with no capping. This
means that the influence of the interface layers that has such a great impact in
the thin film case is reduced for such nanowires. Two Pt electrodes are directly
written onto the crystalline as-deposited nanowires which can then undergo the
SET and RESET operations many times. So nanowires are not only an impor-
tant system for investigating crystallization at a nanoscale size, but they can
also be directly employed as memory devices, even if a new device architec-
ture needs to be developed [81]. Electrical characterizations of reamorphized
nanowires show a decrease of the recrystallization time at fixed temperature,
an increase in the nucleation rate and a decrease in the activation energy for
decreasing diameter (Figures 3.6a, b and c, respectively).
73
3.1 Introduction
Figure 3.6: Measured values for the (a) recrystallization time at fixed temperature, (b)
nucleation rate and (c) activation energy as a function of nanowires diameter as reported in
Ref.[82].
74
3.1 Introduction
The high surface-to-volume ratio of nanowires seems to lead to enhanced
heterogeneous nucleation that can be responsible for the observed size-depending
effects. These results are in contrast with the increase of the crystallization
temperature with decreasing thickness reported for thin films in section 3.1.1,
but nanowires differ significantly from thin films from many points of view.
Their surface-to volume ratio is different and nanowires of Ref.[81] are uncapped
while a capping layer is usually deposited for thin films. Moreover, thin films
are amorphous as-deposited and it is not possible to reamorphize them after
crystallization. So in the amorphous to crystalline transition for thin films the
starting point can be only an as-deposited amorphous phase. For nanowires,
on the other side, crystallization of a melt-quenched amorphous phase has been
studied, so it may be hazardous to directly compare thin films and nanowires
results.
It is possible to go even further in investigating the effects of confinement
on phase change material by studying PC clusters. Due to the 3D confinement,
clusters are the ideal model for investigating the size effects in memory cells
where both size and interface effects play a role. Indeed, in clusters the interface
effects are enhanced due to the large surface/volume ratio. Moreover, for isolated
clusters in a matrix, the plastic relaxation may be limited by confinement, which
can thus enhance strain effects.
Some attempts to obtain PC nanoclusters have been reported [83, 84, 85,
86, 87, 88, 89, 90]. The techniques used to fabricate the clusters vary from
electron beam lithography to laser ablation and chemical synthesis. For GST
nanoclusters made by electron beam lithography with size range from 20 nm
to 80 nm, no significant change in the amorphous to fcc phase crystallization
temperature has been observed [83]. Smaller clusters (15 nm average diameter)
obtained with diblock copolymer transform directly into the hexagonal phase
[84]. The growth of GST clusters using laser ablation has been reported, with
discrepancies in the reported crystallization behavior. The first results [85] with
large size distribution (from 4 nm to 30 nm) indicate that, for 15 nm size-
selected clusters, amorphous particles with irregular shape are obtained when
the annealing is below 200C , while crystalline particles are observed when the
75
3.1 Introduction
annealing is above 300C . One peculiar observation is the fact that, contrary
to bulk or thin films, the hexagonal and fcc phases are formed when annealing is
performed at 300C whereas the pure fcc phase is only observed when annealing
is performed above 400C . A second report, with similar size distribution,
indicates clusters with a mixture of amorphous and fcc phase for all temperatures
from 100C up to 500C [86]. Using a similar growth technique, a third report
indicates mostly amorphous as-prepared particles transforming into a mixture
of hexagonal and fcc phase after annealing at 100C , and in a mostly fcc phase
after annealing at 200C [87].
The main conclusion from these studies is that both the cubic fcc and amor-
phous phases can be observed in nanometric GST clusters. However, the obser-
vation of an unambiguous phase transition from an amorphous to a crystalline
structure, at a definite temperature, has not yet been achieved. Moreover, in
these studies the effect of the cladding material on the phase change proper-
ties of clusters has not been addressed. In order to investigate these aspects,
GST clusters with size below 10 nm and small size distribution are needed.
In the case of GeTe nice results have been obtained using chemical synthesis
[88, 89, 90]. Small particles with size ranging from 1.75 to 3 nm show a Tx more
than 150C above that of the bulk, [89] while the increase for 8.7 nm clusters is
only 67C [90]. However, results are difficult to compare to thin films measure-
ments due to the different fabrication methods and to the possibility of carbon
contamination of the nanoparticles which could contribute to the observed in-
crease of Tx (see Chapter 2). The chemical synthesis of GST clusters has not
been reported up to now. No consistent information have been obtained up to
now on the crystallization temperature of PC nanoparticles beyond 10 nm of
diameter synthesized by sputtering, which is the deposition method commonly
used for thin films deposition.
In conclusion, information in literature about size effect on PC materials
rely on studies of thin films, nanowires and clusters. Thin films show a strong
dependence of the crystallization mechanism on the interface materials and often
exhibit an increase of Tx with decreasing thickness, while nanowires free from
capping materials show a decreasing Tx with decreasing diameter. However,
76
3.2 Clusters deposition
as explained above, it is difficult to interpret the differences observed between
films and nanowires without further studies, for example on capped nanowires.
Nanoparticles are characterized by a high surface to volume ratio and by a strong
influence of capping layers. In the following section the use of a new method
for the growth of GST nanometric clusters by sputtering (section 3.2) and their
characterization through X-ray diffraction (section 3.3) will be described.
3.2 Clusters deposition
In this section the fabrication and deposition of GST nanometric clusters with
average size below 10 nm, embedded in alumina, is described. The process has
been done by R. Morel and A. Brenac (INAC/SP2M, CEA Grenoble) [91]. The
clusters are prepared in a UHV chamber equipped with a sputtering-gas phase
condensation cluster source and two additional standard sputtering guns. The
deposition apparatus is depicted in Fig.3.7. The cluster source itself consists in
a magnetron sputtering head installed in a liquid nitrogen cooled insert. The
stoichiometric Ge2Sb2Te5 solid target is DC sputtered in a 0.1 mbar cold argon
atmosphere, which makes the sputtered vapor condensate into nanometer-sized
clusters. The clusters drift along the gas flow lines and are expelled through
an iris diaphragm in the vacuum, forming a beam which is directed onto the
sample in a deposition chamber. This chamber is equipped with an additional
magnetron that is used for the sputter deposition of GST thin films (using the
same target as the one used for clusters), and a second one for the deposition
of Al2O3 underlayers and capping layers. All depositions are made at room
temperature. The clusters size distribution, given by a time of flight mass spec-
trometer (TOF) , is shown in Figure 3.8a. The average cluster size is 5.7 nm,
and the width of the distribution is ± 1 nm at half maximum. The morphology
of the clusters was controlled by Transmission Electron Microscopy (TEM): a
low density layer of clusters was deposited on an ultra-thin carbon grid, and pro-
tected by a 1 nm Al2O3 layer. The TEM image is reported in Figure 3.8b. The
distribution of the clusters on the surface is random, as expected from the de-
position technique. The particles are spherical. The fact that no atomic planes
77
3.2 Clusters deposition
Figure 3.7: Schematic drawing of the apparatus used to deposit the nanocluster samples
as long as the thin film samples used for comparison. The deposition procedure is briefly
illustrated.
78
3.2 Clusters deposition
Figure 3.8: (a) TOF size distribution which shows that the nanoclusters have an average
diameter of 5.7 nm with a narrow size distribution (± 1 nm at half maximum) (b) TEM images
of GST as-deposited clusters which indicate that the particles are spherical and amorphous.
These images have been made on a low density dedicated sample and the clusters state and
shape have been checked over clusters with the largest diameter. (By courtesy of M. Audier,
LMGP CNRS, Grenoble INP-Minatec
79
3.2 Clusters deposition
Figure 3.9: Scanning Electron Microscopy (SEM) image of GST deposited clusters. The red
circle has a diameter of 20 nm. No trace of particles coalescence can be seen.
are visible for the clusters and that the contrast is similar for all particles is a
first indication that the as-deposited clusters are amorphous, as will be demon-
strated with the X-ray diffraction analysis. In Figure 3.9 a Scanning Electron
Microscopy (SEM) image of the same sample shows no sign of coalescence of the
deposited particles.
Two types of GST samples have been prepared in order to perform the X-ray
diffraction (XRD) measurements. The GST clusters samples consist in 4 GST
clusters layers deposited on Si. First a layer of 6 nm of Al2O3 has been deposited
on the Si substrate. Then a layer of GST clusters with an equivalent mass of
0.07 monolayer of particles has been deposited, covered with a 3 nm Al2O3 layer.
Then, three other layers of GST clusters have been deposited, each one covered
by 3 nm of Al2O3 , and at the end a 6 nm Al2O3 layer has been deposited. The
final structure, shown in Fig. 3.10a, is 6 nm Al2O3 / (GST clusters layer / 3
nm Al2O3 ) × 4 / 6 nm Al2O3 . The average distance between clusters is two
cluster diameters. This structure has been chosen in order to avoid sintering
effects during annealing. The GST film samples consist in 10 nm GST thin film
sandwiched between two 10 nm thin Al2O3 films, deposited on a Si substrate.
80
3.3 X-Ray Diffraction study
Figure 3.10: (a) GST clusters sample: 4 layers of clusters capped with Al2O3 are deposited
on an Al2O3 substrate and capped again with Al2O3 . The average distance between clusters
in a layer is about 2 cluster diameters. (b) GST film sample: 10 nm thick film of GST
sandwiched between two 10 nm thin Al2O3 films. Both the clusters and film samples are
deposited on a substrate of Si.
Their schematic representation is reported in Figure 3.10b. The GST quantity
in clusters samples is five times less than the one in film samples. A 20 nm
Al2O3 thin film on a Si substrate, hereafter called blank sample, was also grown
for background signal measurement in the X-ray diffraction experiments. Thin
films and clusters compositions were measured with Rutherford Backscattering
Spectroscopy (RBS) and Particle-Induced X-ray Emission (PIXE). For thin films
the content is Ge:Sb:Te = 23:24:53 (±3), very close to the Ge2Sb2Te5 = 22:22:56
target composition. For the clusters the composition is 28:27:45 (±3), which
indicates a slight Te depletion as often reported for very thin films and clusters
[70, 83, 86, 92]. Within experimental accuracy, the GST clusters composition is
found identical before and after annealing.
3.3 X-Ray Diffraction study
In order to qualitatively locate Tx and observe the crystallized state in clus-
ters a first set of clusters samples was annealed under vacuum (10( − 3) bar) at
200C and 300C , while a film sample was annealed at 200C . The tempera-
81
3.3 X-Ray Diffraction study
ture was ramped at + 10C /min, held at the set point temperature for 30 min,
and ramped down to ambient temperature. X-ray diffraction on those samples,
hereafter referred to as ex situ samples, has been performed at room tempera-
ture. For a second set of clusters and thin films samples the X-ray diffraction
spectra were recorded as a function of the temperature, during the annealing
process. Those samples will be referred to as in situ samples. X-ray diffraction
measurements were performed using synchrotron radiation on the BM02 CRG-
D2AM beamline (ESRF Grenoble, France) with a photon incident energy of 17.8
keV (λ =0.69654 A) and using a 2D CCD camera detector. The detailed de-
scription of the experimental setup is reported in appendix A.2.2. The incident
angle was 4 and the position of the CCD camera allowed for measurements in a
2θ range from 8 to 26. As it is reported in details in appendix A.2.2 about the
data treatment description, the contribution of the Si substrate and deposited
Al2O3 are very intense compared to the GST signal. These background con-
tributions need to be subtracted from the measured diffracted signal. For this
purpose, the blank sample (Si + Al2O3 ) has been measured. The 2D image
obtained for the as-deposited GST film sample, after removing the background,
shows only faint traces, with no diffraction rings. This is a first confirmation that
the as-deposited GST film is amorphous. The 2D image of the 200C ex situ
annealed thin film sample, again after background subtraction, shows diffrac-
tion rings which can be indexed as a fcc structure with a (111) texture (Figure
3.11a).
Concerning the as-deposited clusters, the X-ray diffraction 2D image shows
no diffraction rings. For the 200C annealed clusters, supposing a cubic struc-
ture, the (111) and (222) diffraction rings are not observed but a weak signal
is visible for the (200) and (220) diffraction rings (Figure 3.11b), which are the
two most intense reflections expected in a powder diffraction pattern from the
bulk fcc phase. The intensity is constant over all the measured rings portions.
A diffuse scattering from the Si substrate is still present at the corners of image
Fig.3.11b, showing that the background subtraction is not perfect1.
1This could be related to the fact that the relative contributions of the Al2O3 signal and
Si signal are not the same in the blank sample and in presence of GST. This problem could
82
3.3 X-Ray Diffraction study
Figure 3.11: (a) X-ray 2D diffraction images for 200C ex situ annealed GST thin film. (b)
Same measurements for 200C ex situ annealed clusters. In both cases, the 2D image of the
blank sample (Si+Al2O3 ) has been subtracted.
The 2D images give information about the samples texture and traces of
Bragg peaks can be detected directly on them, but the evidence of the phase
transformation and the position for the diffraction lines are more clearly ob-
served by measuring the angular integrated intensity, in particular for clusters.
The intensity as a function of 2θ has been obtained by integrating over the en-
tire 2D images for each sample, excluding only a border of 100 pixels on each
side, thus losing information on texture in the film case. The background inte-
grated intensity has always been subtracted. The angular integrated intensity
for the film sample is shown in Figure 3.12. For the as-deposited film the clear
observation of two broad maxima at 2θ = 12.9 and 21.6 allows to conclude
that as-deposited clusters are amorphous. Their positions, corresponding to
Q = 4πsinθ/λ =2.03A−1 and 3.38 A−1, match the first two maxima of the GST
amorphous structure factor reported in literature [27] (see Chapter 2). For the
annealed film the peaks positions indexed in the fcc cubic structure indicate a
lattice parameter of 6.01 ± 0.01 A that closely matches the one reported for
bulk GST, 6.0117(5) A [25]. It should be noticed that due to the texture and
the finite angular range the relative integrated intensity for the peaks is not the
one expected for a powder pattern. Such intensity is reported in Table 3.2 as
calculated considering the lattice parameter for GST given in Ref. [25].
not be solved after the experience took place.
83
3.3 X-Ray Diffraction study
8 10 12 14 16 18 20 22 24
0
1x107
2x107
3x107
4x107
220
200
222
111
As-deposited
200°C
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
2q[°]
311
Figure 3.12: X-ray diffraction spectra at room temperature for as-deposited and 200C ex
situ annealed GST film, after background subtraction, with curves shifted for clarity. Ar-
rows indicate bulk GST fcc peak positions calculated assuming the lattice parameter of GST
a=6.0117 A reported in Ref.[25].
84
3.3 X-Ray Diffraction study
12 14 16 18 20 22
0.0
2.0x106
4.0x106
6.0x106
8.0x106
1.0x107
1.2x107
220200
As-deposited
200°C
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
2q[°]
Figure 3.13: X-ray diffraction spectra at room temperature for as-deposited and 200C ex
situ annealed GST clusters after background subtraction. Curves are shifted for clarity. Ar-
rows indicate bulk GST fcc peak positions calculated assuming the lattice parameter of GST
a=6.0117 A reported in Ref.[25].
85
3.3 X-Ray Diffraction study
hkl 2θ[] d[A] Irel
111 11.518 3.47086 4.25
200 13.307 3.00585 61.18
220 18.862 2.12546 100.00
311 22.155 1.81260 8.62
222 23.154 1.73543 57.35
Table 3.2: Peak positions and relative intensities for the fcc GST phase as expected from a
powder pattern. They have been estimated in a θ −2θ geometry at the actual experimental
wavelength considering the lattice parameter a=6.0117 A reported in Ref [25].
The angular integrated intensity obtained for ex situ clusters samples is re-
ported in Fig. 3.13. The as-deposited sample shows no clear trace of the amor-
phous broad maxima that can be identified for the as-deposited film sample,
as can be understood in view of the smaller quantity of GST in clusters sam-
ples. As expected, due to the smaller crystallite size, the peaks width for the
200C annealed cluster samples are larger than for the thin film. Besides, a clear
shift of the diffraction lines with respect to their position in the crystalline thin
film is observed.The fcc lattice parameter for the crystalline clusters calculated
from the (200) and (220) peaks position is 6.11 A ± 0.02.
In order to measure Tx more precisely the 2D X-ray diffraction spectra for
as-deposited thin films and clusters were recorded as a function of the tempe-
rature, with in situ annealing using a domed oven stage. More details on the
experimental setup and data treatment can be found in appendix A.2.2. These
measurements are challenging due to the temperature dependent spurious signal
from the oven dome that reaches more than 109 counts, two orders of magnitude
higher than the film signal and even three orders of magnitude higher than the
clusters signal. Nevertheless, the crystallization of the amorphous as-deposited
clusters could be observed. The temperature was increased by 10C steps after
which X-ray diffraction spectra were recorded for 26 min.
The change in the thin film (220) diffraction peaks from 150C to 180C is
plotted in Figure 3.14. Despite the deterioration of the signal over noise ratio
in presence of the dome covering the oven, it is clear that the peak intensity
86
3.3 X-Ray Diffraction study
18.0 18.2 18.4 18.6 18.8 19.0 19.2 19.4
0.0
5.0x106
1.0x107
1.5x107
2.0x107
Peak 220
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
2q[°]
140°C
150°C
160°C
170°C
180°C
200°C ex situ
220
Figure 3.14: (220) diffraction peak for in situ annealed GST thin film at different tempera-
tures. The dotted line is the peak, measured at room temperature, of the thin film annealed ex
situ at 200C (shifted for clarity). The arrow indicates the calculated bulk GST peak position.
87
3.3 X-Ray Diffraction study
12.5 13.0 13.5 14.0 14.5
0
1x106
2x106
3x106
4x106
5x106
17.5 18.0 18.5 19.0 19.5 20.0
(a)ex situPeak 200
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
2q[°]
150°C
170°C
190°C
210°C
230°C
(b)ex situ
Peak 220
2q[°]
220200
Figure 3.15: X-ray diffraction spectra for in situ annealed GST clusters at different temper-
atures. (a) (200) diffraction peak and (b) (220) diffraction peak. Curves are evenly shifted to
ease viewing. Dotted lines indicate the peak position, measured at room temperature, of the
200C ex situ annealed clusters.
measured at 180C is comparable with that measured at room temperature for
the sample annealed ex situ at 200C . At 150C no signal is recorded above
background level, while at 170C the crystalline peak is close to its final ampli-
tude. It can be observed that the (220) peak position at 180C is slightly below
that of the ex situ annealed thin film measured at room temperature, which can
be explained by the GST thermal dilatation [93, 94]. Considering a coefficient
of thermal expansion for Ge2Sb2Te5 of αT = 1.81 ·10−5K−1 [94] and the relation
d = d0(1 + αT ·∆T) for the thermal expansion of the interatomic distances, 2θ
is expected to vary of 0.05 between the room temperature and 200C .
The (200) and (220) diffraction peaks for in situ annealed clusters are shown
in Figure 3.15. Despite the high noise level and remaining contribution from the
oven dome at 2θ=19.5(Fig. 3.15b), the parallel rise in the amplitude for the
two peaks is visible. In order to make a more quantitative analysis the area of
the peaks has been integrated as a function of temperature2. Considering the
2Integration is restricted to the left part of the 220 peak in order to avoid the remaining
spurious signal of the dome.
88
3.3 X-Ray Diffraction study
Figure 3.16: Normalized integrated intensities for (220) and (200) diffraction peaks for GST
clusters and for GST film as a function of temperature. The normalized integrated intensities
have been obtained through the relation Inorm = I/Imax where I is the measured integrated
intensity at a given temperature and Imax is the maximum value of the integrated intensity
(which correspond to complete crystallization). The dotted lines indicate the crystallization
temperatures.
different values of the peak maxima for film and clusters samples, the obtained
integrated intensities have been normalized through the relation Inorm = I/Imax
where Inorm is the normalized integrated intensity and Imax is the maximum
value of the integrated intensity (which corresponds to complete crystallization).
All these intensity are measured at a given temperature. Inorm corresponds to
the crystalline fraction, thus allowing a direct and clearer comparison between
clusters and film. The crystallization temperature is defined as the temperature
corresponding to the midpoint of the rise step of Inorm. The results are shown
in Figure 3.16.
For the thin film the rise is almost parallel for the (220) and the (200) peaks.
No crystallization occurs at temperature lower than 140C . The value of Tx is
155C and the crystallization is completed at 170C . For the clusters, crystal-
lization starts at around 150C , Tx is close to 180C and the rise in amplitude
is more gradual, spanning from 150C up to 200C .
89
3.4 Discussion
3.4 Discussion
The most important result obtained from the ex situ cluster samples is the
confirmation that the nanoparticles are able to switch from the amorphous to
the crystalline fcc phase when annealed. So, even for particles with a diameter
as small as 5 nm, the phase change mechanism still takes place. This result is
positive for the future developments of Phase Change Memories.
A second key observation regarding the annealed crystalline clusters is that
the lattice parameter obtained from the measurements is larger than that of
the annealed fcc thin films, the latter being very close to the one expected for
the GST fcc cubic phase. For thin films, Scherrer analysis gives a lower limit
for the crystallite size [95] close to the layer thickness and a small inhomoge-
neous strain of 0.006. These films are close to complete relaxation. On the
other hand, the shift in the crystalline clusters peaks position indicates that the
lattice parameter for the fcc clusters (6.11A) is 1.7% larger than that of the
crystallized thin film (6.01A), which can be attributed to a large tensile strain
due to the interaction with the oxide matrix3. At the phase change transition,
the GST density decreases by 5% [25, 94]. In thin films, the stress resulting from
the volume change can be relaxed along the two space directions that are not
stuck to the surrounding layers, whereas clusters are bound to the surrounding
matrix in all three dimensions. Upon crystallization, the deformation must be
accommodated one way or another. This can be for instance via the creation
of voids [96, 97] or vacancies [98] or strain. In the present case, the observed
homogeneous tensile strain is very close in absolute value to the reduction in size
that would be expected for freestanding clusters during the amorphous to crys-
talline phase change. Indeed, the variation of the lattice parameter estimated for
clusters compared to the thin film ∆ aa
≈ 1.7% corresponds to a relative volume
variation ∆ VV
= (∆ aa)3 ≈ 4.9%. This suggests the scenario represented schemat-
ically in Fig.3.17. When an amorphous cluster switches to the crystalline phase
the strong interaction with the embedding alumina, which is a far more rigid
3In crystalline clusters, there are only two peaks that can be analysed. A Sherrer analysis
with no inhomogeneous strain leads to a grain size in agreement with the clusters diameter.
Therefore, inhomogeneous strain effects in clusters will not be considered.
90
3.4 Discussion
Figure 3.17: The Al2O3 matrix surrounding the cluster forces its volume to remain equal
to the one of the amorphous phase even after crystallization. The cluster in its amorphous
phase occupies a certain volume (a). When crystallization occurs, the cluster volume tends
to reduce of around 5% (b), but the embedding Al2O3 matrix exerts a tensile strain over the
cluster (c) thus forcing the cluster to keep the volume corresponding to the amorphous phase,
with the effect of increasing the lattice parameter of the crystalline cluster (d).
material than GST, forces it to keep the volume that it had in the amorphous
phase.
As already stated, there are no clear reported values of the crystallization
temperature for GST nanoclusters of such small diameter in literature. Never-
theless, it seems quite natural to compare our results with those reported in Ref.
[70] on GST thin films embedded in Al2O3 , which is the same cladding mate-
rial used in this study. The Tx obtained for the thin film in this work (155C )
is in agreement with the value reported in Ref.[70] for a GST film of similar
thickness. On the other hand, the Tx in the case of clusters is only 25C above
Tx of the 10 nm thick GST thin film, an effect much smaller than reported in
Ref.[70] where a thin film of 5 nm shows a Tx of more than 330C . However, the
different surface to volume ratio can have a deep influence on the crystallization
mechanism, making difficult a direct comparison between clusters and thin films
scaling properties. Moreover, a film is free to change its volume upon phase
transformation while clusters are confined in three dimensions and strained, as
shown above. The strain may possibly play a role in promoting or impeding
crystallization. However, no data about the deformation or strain of measured
films are available in Ref. [70], thus impeding to compare strain effects. In Ref.
[72] an effect of stress over crystallization has been addressed, but the variation
on lattice parameters due to strain has not been measured so again a comparison
is hazardous.
91
3.4 Discussion
As shown in section 3.1.1, it is very difficult to discuss effects on the crys-
tallization temperature since a wide number of factors plays a role. However,
considering that in the present case the GST clusters, the GST film and the
Al2O3 capping have been deposited in the same apparatus and measured in
identical conditions it is possible to try to interpret the observed 25C variation
of Tx between thin film and clusters. It could be related to many different factors
including composition effect, different surface to volume ratio, matrix influence,
stress or strain effects or an intrinsic size effect. A possible composition effect
could arise from the fact that, as compared with the films, the clusters are Te-
depleted. For instance the crystallization temperature in Ge-Sb-Te thin films
with 10%-20% excess Sb is 15C higher than for GST films [99]. The crystal-
lization temperature for Ge2Sb2Te4 is 175C [100].Another effect could be the
stress resulting from the phase change. From a thermodynamical point of view
the elastic energy stored in clusters will reduce the driving force for the phase
transition. The kinetics for the phase change will be slowed and, during a tem-
perature scan, the transition temperature will increase. In the case of GST this
driving force is 200 MJ/m3, [101]. An order of magnitude for the elastic energy
resulting from the strain is close to 50 MJ/m3, considering the experimental
value of the bulk modulus available in literature [102], so it can induce signifi-
cant effects. Finally, the increase of Tx could be an intrinsic size effect due to
the impact of surface energy as explained through Zacharias model [71], even
if the effect is much smaller than what reported in Ref.[70] for GST thin films
capped with Al2O3 .
As already mentioned at the end of section 3.3, the crystallization of clusters
is more gradual than the one of the 10 nm film. This can be due to a disper-
sion of the crystallization temperatures of clusters. If Tx increases for reduced
dimensions due to a size effect, smaller clusters will have a higher Tx so the
crystallization temperature dispersion is a consequence of the size distribution.
However, it is also possible to have a dispersion of Tx even for clusters of the
same size, due for example to oxidation of the clusters at the Al2O3 interface
which can change the composition and influence Tx . It is also worth noting
that, supposing an homogeneous nucleation and considering that the clusters
92
3.5 Conclusions and perspectives
size is quite close to the critical nucleus diameter (about 2 nm), a dispersion of
Tx could occur from cluster to cluster.
3.5 Conclusions and perspectives
In conclusion, nanometric GST clusters were deposited by a sputtering gas-
aggregation technique, with a narrow size distribution around 5.7 nm. The
obtained results demonstrate that this synthesis technique offers new possibility
for the study of well calibrated and isolated clusters of phase change materials,
opening the way to a systematic study of nanoparticles deposited by sputtering.
It offers the same reduced dimensions and size distribution as chemical synthe-
sis but with the possibility of depositing ternary compounds (this is still very
difficult with chemical techniques) and with a much smaller risk of including
contaminants. Moreover, this method is close to the physical techniques used
for PC film deposition in the fabrication of PCM, thus giving information that
can be more directly exported to device fabrication. The as-grown clusters dis-
persed in alumina are amorphous and transform into a fcc crystalline phase at a
well defined crystallization temperature of 180C . This is the first unambiguous
observation of this phase change in GST clusters in the sub-10 nm range. The
crystalline clusters show a lattice parameter larger than that of bulk cubic GST,
which indicates a tensile stress that can be attributed to the interaction with the
alumina matrix. The large increase in Tx observed in thin GST films subjected
to large interface stress [72] is not seen in clusters. These results indicate that
the scaling effect on the crystallization temperature in phase change material
can be small and the role of interfaces in term of stress and interfacial energy
effects must be further studied.
All the results reported in this section have been obtained from a single
experience at the synchrotron that was done over the first set of GST clusters
samples obtained with this deposition method. This experience allowed to learn
how to improve the data acquisition and subsequent analysis. For the next
experiences the quantity of matter in clusters samples should be increased in
order to obtain more intense signals, improve the quality of the data and reduce
93
3.5 Conclusions and perspectives
the effort for the data treatment that was difficult and time consuming. The
blank sample needs to be deposited with the same amount of cladding material as
the PC samples in order to avoid errors in subtracting the substrate contribution.
In the near future the same deposition technique will be used to deposit
GeTe nanoparticles. Besides, based on the results presented in Chapter 4, the
effect of different capping layers as Ta, TiN, W or SiO2 will be also investigated.
Smaller clusters could also be deposited, but their size distribution would be
larger. A further step will be the electrical characterization of clusters in order
to determine their electrical properties such as threshold voltage, retention time,
reset current and cyclability. Doped clusters can also be deposited in order to
study the effect of doping at small sizes, and to evaluate the impact of the
variability of doping concentration in such small systems. To go further, local
order investigations and TEM measurements can be done in addition to XRD
measurements in order to completely characterize the structure of clusters.
94
Chapter 4
Interface effect on crystallization
of PC thin films
95
4.1 Introduction
4.1 Introduction
As already mentioned in section 3.1.1, the material used as a capping layer
has been shown to have an influence on the crystallization mechanism of a
phase change very thin film. Even for films as thick as 30 nm the same Ge-
Sb-Te compound tested under the same conditions crystallizes at a different
temperature if interfaced with different materials such as silicon dioxide SiO2 ,
silicon nitride Si3N4, tantalum oxide Ta2O5, zinc sulfide ZnS or ZnS − SiO2
[66], or, even when the value of Tx is unchanged, the crystalline phase and
growth rates of the same PC material can be different if capped with Si3N4 or
ZnS − SiO2 [77]. For film of GST and GeTe of larger thickness, the effect of
interfaces is assumed to be absent and the measured Tx is considered as the
bulk crystallization temperature.
In this chapter a study on the effect of different interfaces on the crystal-
lization temperature of a PC material thin film will be reported. The value
of Tx will be measured through reflectivity measurements (section A.1) and X-
Ray Diffraction measurements (XRD) (sections 4.3 and 4.4), and a possible
interpretation of the results will be proposed at the end of the chapter (section
4.5). The samples tested though reflectivity consist in thin films of GeTe and
GST of various thicknesses encapsulated in three different materials, TiN, Ta or
SiO2 . Those test materials have been chosen as capping layers because they are
frequently employed in the fabrication process of devices (TiN and Ta for the
electrodes and SiO2 as insulating material), so they can be eventually integrated
in a cell. The structure of the samples, shown in Fig. 4.1, is the same for both
GeTe and GST samples. The XRD experiments have been performed on GeTe
thin films 100 nm thick only. All the phase change materials and capping layers
have been deposited by sputtering as described in appendix B. The measured
composition of the GeTe samples is Ge52Te48. The experimental procedures and
setups are described in details in appendix A.2.
96
4.2 Reflectivity measurements
!"#
!$# !%#
!"#$%&$'()'*
+",#-.#/0
!"#$%&$'()'*
12#0)'*(")3
+",#4#/0
!"5-#4..#/0 +)#-.#/0
!"#$%&$'()'*
12#0)'*(")3
+)#4#/0
12#0)'*(")3
!"5-#6.#/0
Figure 4.1: Structure of PC material thin films samples used in order to investigate the
interface effect on crystallization. The PC material can be either GeTe or GST, of various
thicknesses, sandwiched between (a) SiO2 , (b) TiN or (c) Ta. All the samples have been
deposited by sputtering as described in B.
4.2 Reflectivity measurements
The reflectivity as a function of temperature of 100 nm thin films of GST and
GeTe sandwiched between TiN, Ta or SiO2 is shown in Figure 4.2. The heating
rate was 10C /min for all the samples. The amorphous to crystalline phase
transition corresponds to an increase in the value of the measured reflectivity.
It is evident from Fig. 4.2 that for both GST and GeTe the phase change
occurs at similar temperatures when the film is sandwiched between TiN or
SiO2 , while the transition occurs at a higher temperature when the PC material
is interfaced with Ta. This is a surprising result. Even if an effect of the
encapsulating material on Tx has been observed in literature for films of 30 nm
[66] it was very small, while in the present case we observe a difference of more
than 20C even for films 100 nm thick. The crystallization temperatures Tx for
the six samples of Fig. 4.2, calculated as the point of maximum derivative
of the reflectivity curve, are reported in Table 4.1. The difference between
the crystallization temperatures obtained with the TiN or SiO2 interfaces is
small both for GeTe and GST samples, and it is within the uncertainties on
the sample temperature in the reflectometer 1. The resulting values of Tx for
GST interfaced with SiO2 and TiN are close to the Tx reported in literature for
1As described in details in appendix A.1, the temperature in the reflectometer is measured
on the heating plate. If the thermal contact between the heating plate and the sample is not
perfect the actual temperature on the sample can differ from the measured one.
97
4.2 Reflectivity measurements
150 160 170 180 190 200 210 220 230
0.0
0.2
0.4
0.6
0.8
1.0
120 130 140 150 160 170 180 190 200
0.0
0.2
0.4
0.6
0.8
1.0
(a)
Cry
sta
llin
e f
ract
ion
GeTe 100 nm / SiO2
GeTe 100 nm / TiN
GeTe 100 nm / Ta
(b)
Temperature [°C]
GST 100 nm /SiO2
GST 100 nm / TiN
GST 100 nm / Ta
Figure 4.2: Crystalline fraction as a function of temperature obtained from reflectivity
measurements for (a) GeTe and (b) GST 100 nm thin films sandwiched between TiN, Ta or
SiO2 heated at 10C /min. For both GeTe and GST thin films the amorphous to crystalline
transition occurs at a higher temperature when the film is sandwiched between Ta.
98
4.2 Reflectivity measurements
GeTe 100 nm Tx [C ] GST 100 nm Tx [C ]
SiO2 188 148
TiN 193 152
Ta 213 165
Table 4.1: Crystallization temperatures Tx of GeTe and GST 100 nm thick films sandwiched
between TiN, Ta or SiO2 as obtained from the reflectivity measurements of Fig. 4.2 for a
heating rate of 10C /min.
30 nm thick films of GST interfaced with SiO2 and ZnS-SiO2 characterized by
transmittance measurements and resistance measurements [66, 67]. They are
lower than what reported for TiN interfaced GST measured by EXAFS and
ellipsometry in Ref.[72] and for GST interfaced with Al2O3 and measured by
X-Ray Diffraction [70], but in this last case the heating rate is six times higher
than in our experiments. The values of Tx obtained for GeTe films interfaced
with SiO2 and TiN are in agreement with what reported in Refs.[23] and [44] for
thin films measured under the same condition as in this work at a heating rate
of 20C /min and 10C /min, respectively. A Tx of 175C has been reported
in Ref.[103] for resistance measurements on a GeTe film 50 nm thick interfaced
with SiO2 heated at 1C /s and a Tx of 170C has been reported in Ref.[104]
for a GeTe film 75 nm thick interfaced with polymethyl methacrylate heated at
23C /min and measured by optical transmission measurements.
The same reflectivity measurement has been repeated on samples of GeTe
30 nm and 10 nm thick interfaced with TiN, Ta or SiO2 and the results are
shown in Fig.4.3. For the 30 nm thick samples the change in reflectivity can
be easily identified for each kind of interface, and the difference in Tx observed
for the 100 nm thick films between the Ta interfaced samples and the others is
still clearly evident. For the 10 nm thick GeTe films the phase transition for the
SiO2 interfaced sample is still evident, but the interpretation of the reflectivity
measurement becomes difficult for the TiN and Ta interfaced samples due to the
high contribution of the upper layer of TiN or Ta, that could not be subtracted.
However, a phase change can still be observed for the TiN interfaced sample.
On the other hand, no reflectivity change could be detected for the Ta interfaced
99
4.2 Reflectivity measurements
GeTe 30 nm Tx [C ] GeTe 10 nm Tx [C ]
SiO2 191 197
TiN 187 211
Ta 223 >230
Table 4.2: Crystallization temperatures Tx of GeTe films 30 nm and 10 nm thick sandwiched
between TiN, Ta or SiO2 as obtained from the reflectivity measurements of Fig. 4.3 for a
heating rate of 10C /min.
sample in the experimental temperature range up to 230C . It could be due to
the fact that Tx for that sample is above 230C , but no definitive conclusion
can be drawn in view of the difficulties of the reflectivity measurements. The
values of Tx for the 30 nm and 10 nm thick GeTe samples are shown in Table 4.2.
For the 30 nm thick films, the difference of Tx between the Ta interfaced sample
and the others is around 30-35C , while it was around 20-25C for 100 nm thick
GeTe films. By comparing Tables 4.1 and 4.2 it can be concluded that reducing
the film thickness has a weak effect on Tx for the SiO2 interfaced samples, while
Tx seems to increase with deacreasing thickness for the samples interfaced with
TiN and Ta. Indeed, the 100 nm and 30 nm TiN interfaced samples have the
same Tx , while Tx increases of around 20C for the 10 nm thick film. The Tx of
Ta interfaced GeTe samples increases of around 10C when the film thickness is
reduced to 30 nm and no phase transition can be seen up to 230C for the GeTe
film 10 nm thick.
The activation energy EA of the GeTe 100 thin film has been calculated for
the SiO2 and Ta interfaced samples by repeating the reflectivity measurement
for three other heating rates: 3C /min, 5C /min and 20C /min. The crystal-
lization temperatures Tx obtained for all different heating rates are reported in
Table 4.3. The values of EA obtained by fitting the set of points shown in Fig.
4.4 (as described in section 1.3.5) are of 3.83 eV for Ta interface and 2.58 eV for
SiO2 interface. The activation energy obtained for GeTe interfaced with SiO2 is
close to the values known from literature [23]. EA is significatively enhanced for
the Ta interfaced sample.
100
4.2 Reflectivity measurements
150 160 170 180 190 200 210 220 230
0.0
0.2
0.4
0.6
0.8
1.0
150 160 170 180 190 200 210 220 230
0.0
0.2
0.4
0.6
0.8
1.0
(a)
Cry
sta
llin
e f
ract
ion
GeTe 30 nm / SiO2
GeTe 30 nm / TiN
GeTe 30 nm / Ta
(b)
Temperature [°C]
GeTe 10 nm / SiO2
GeTe 10 nm / TiN
GeTe 10 nm / Ta
Figure 4.3: Crystalline fraction as a function of temperature from reflectivity measurements
for (a) GeTe 30 nm and (b) GeTe 10 nm thin films sandwiched between TiN, Ta or SiO2 ,
heated at 10C /min. For 30 nm thick GeTe films the amorphous to crystalline phase transition
occurs clearly at a higher temperature when the PC material is sandwiched between Ta. For
10 nm thick films of GeTe the measurement becomes difficult for samples interfaced with TiN
and Ta, while Tx can be still easily identified for the SiO2 interfaced sample.
101
4.2 Reflectivity measurements
23.6 23.8 24.0 24.2 24.4 24.6 24.8 25.0 25.2 25.4 25.6 25.8
-11.5
-11.0
-10.5
-10.0
-9.5
-9.0
EA=2.58 eV
SiO2 interface
Ta interface
-ln
(r/T
x²)
1/(kBT
x) [eV
-1]
EA=3.83 eV
Figure 4.4: Kissinger plot for GST 100 nm thin films sandwiched between Ta or SiO2 . The
absolute value of the line slope corresponds to the activation energy EA . The points in graph
have been calculated from the Tx value obtained for four different heating rates r as reported
in Table 4.3
102
4.3 X-Ray Diffraction measurements
Heating rate Tx for SiO2 interface [C ] Tx for Ta interface [C ]
3 C /min 179 207
5 C /min 184 210
10 C /min 188 213
20 C /min 191 217
Table 4.3: Crystallization temperature Tx obtained from reflectivity measurements for dif-
ferent heating rates for GeTe 100 nm thin films sandwiched between Ta or SiO2 .
1
2
Detector
SampleBeam
Beam
source
Figure 4.5: Schematic representation of the experimental geometry of the XRD experiment,
where θ is the incident beam angle and Ψ is the tilting angle of the sample. A detailed
description of the experimental setup is provided in appendix A.2.1
4.3 X-Ray Diffraction measurements
Three as deposited 100 nm thick films of GeTe have been characterized by
in situ annealing XRD in order to investigate their crystalline structure. The
samples structure is the same as the one used for the reflectivity measurements
samples (shown in Fig. 4.1). The XRD experiments have been performed at
λ = 1.540598A using a PANalytical instrument equipped with a punctual pixel
detector in θ-2θ configuration as described in A.2.1. The samples have been
annealed under N2 atmosphere from 100C to 300C by steps of 10C . At
each step the temperature was kept constant for 5 minutes before starting the
measurement that lasted 1 hour and 25 minutes. Note that in the reflectivity
measurement the temperature as a function of time is a ramp and the heating
rate is defined as the constant slope of the temperature ramp. For the XRD
experiment the temperature as a function of time is a staircase function. In order
to compare the XRD results and the reflectivity measurements results, it is useful
to define an equivalent heating rate approximated as equal to the step increase
divided by the time step (0.11C /min). Each measurement was performed in
103
4.3 X-Ray Diffraction measurements
25 30 35 40 45 50 55
0
100
200
300
400
500
600
700
800I(
2q)
[#
of
cou
nts
]
2q [°]
SiO2
Ta
TiN
Rhomboedric GeTe
20
2
02
10
06
11
3
01
5
11
0
10
4
10
1
00
3
01
2
Figure 4.6: Diffracted intensity as a function of 2θ for Ψ =0of GeTe 100 nm thin films
sandwiched between TiN, Ta or SiO2 measured at 100C after annealing at 300C with a
heating rate of about 0.11C /min. The vertical lines correspond to the calculated position
of Bragg peaks for rhombohedral GeTe (hexagonal indexation) [25]. No difference in the
diffraction spectra can be observed between the Ta and TiN interfaced samples, while no
peaks are visible for the sample sandwiched in SiO2 . The intense peak around 33 is due to
Ta.
the range 23 < 2θ < 55 for 4 different values of the tilting angle Ψ, shown in
Fig. 4.5, corresponding to 0, 20, 40 and 60. The same measurements have
been done also for a descending temperature ramp after crystallization, from
300C to 100C , under the same conditions for temperature steps of 10C .
The intensity as a function of 2θ for Ψ=0 for the three samples measured at
100C after the annealing at 300C is reported in Fig. 4.6. In the same figure
the peak positions of rhombohedral GeTe at the experimental wavelength are
reported, estimated by considering the lattice parameters a = 4.164 A and c
= 10.69 A with an hexagonal indexation [25]. It is immediately evident that
the SiO2 interfaced sample shows no trace of Bragg peaks, except a weak peak
around 51. Considering that the same sample has been observed to be fully
crystalline after annealing at 230C through reflectivity measurements, the ab-
sence of Bragg peaks is probably an effect of texture, as it will be confirmed
104
4.3 X-Ray Diffraction measurements
in the next paragraph. For the Ta and TiN interfaced samples it is possible to
clearly identify the 101 and 012 peaks, as well as a strongly asymmetric peak
around 43. The peak shape suggests the presence of two peaks that cannot be
distinguished due to the measurement resolution and correspond to the 104 and
110 peaks of the GeTe rhombohedral phase. The intense peak around 33 in the
Ta interfaced film is due to Ta.
In order to check the samples textures at complete crystallization, the 012
peaks of all the samples measured at 100C after annealing at 300C have been
reported in Fig.4.7 for each tilting angle Ψ. It is immediately clear from the figure
that the SiO2 interfaced sample is strongly textured, since the peak intensity
varies strongly with Ψ. In particular, the peak intensity is maximum for Ψ=40,
very weak for Ψ=20 and equal to zero for Ψ=0, indicating that the angle
between the preferred orientation of the diffracting planes and the sample surface
is close to 40. The Ta and TiN interfaced samples show only a weak texture.
The ratio ∆A/Amean, where ∆A is the difference between the maximum and
minimum value of the peaks area and Amean is the mean area over all the values
of Ψ, is around 2.62 for the SiO2 interfaced sample, 1 for the TiN one and 0.5446
for the Ta one, confirming the difference in texture.
Considering that the most intense peaks for the SiO2 interfaced sample are
observed for Ψ=40, the evolution of the 012 Bragg peak of GeTe as a function
of temperature for each sample has been measured for Ψ = 40 and the results
are reported in Fig.4.8. The Bragg peak appears at around 150C for the sam-
ple interface with TiN, 160C for the sample interfaced with SiO2 and 190C for
the sample interfaced with Ta, confirming the difference of crystallization tem-
perature observed in reflectivity measurements (section A.1). The peak area
for each sample has been normalized between 0 and 1 to obtain the crystalline
fraction as a function of temperature, as reported in Fig.4.9, allowing the direct
comparison with the reflectivity data of Fig.4.2. The values of Tx are shown in
Table 4.4.
In the case of in situ XRD experiment for all samples the crystallization takes
place at lower temperatures compared to the reflectivity measurements due to
the different heating rates used in the XRD and reflectivity experiments.
105
4.3 X-Ray Diffraction measurements
28.5 29.0 29.5 30.0 30.5 31.0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
28.5 29.0 29.5 30.0 30.5 31.0
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
28.5 29.0 29.5 30.0 30.5 31.0
0
100
200
300
400
500
600
700
800
900
1000
1100
Psi = 0°
Psi = 20°
Psi = 40°
Psi = 60°
GeTe 100 nm / SiO2 10 nm
2q [°]
2q [°]
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
GeTe 100 nm / TiN 5 nm
Psi = 0°
Psi = 20°
Psi = 40°
Psi = 60°
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
GeTe 100 nm / Ta 5 nm
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
2q [°]
Psi = 0°
Psi = 20°
Psi = 40°
Psi = 60°
Figure 4.7: 012 Bragg peak for 100 nm thick GeTe films interfaced with SiO2 , TiN and Ta
measured at 100C after annealing at 230C (heating rate of about 0.11C /min) for various
tilting angles Ψ of the samples. The Ta and TiN interfaced samples show a weak texture
while the sample sandwiched in SiO2 is strongly textured, with a maximum peak intensity for
Ψ=40 and no intensity for Ψ=0.
106
4.3 X-Ray Diffraction measurements
28.0 28.5 29.0 29.5 30.0 30.5 31.0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
28.0 28.5 29.0 29.5 30.0 30.5 31.0
0
100
200
300
400
500
600
700
800
900
1000
28.0 28.5 29.0 29.5 30.0 30.5 31.0
0
100
200
300
400
500
600
700
800
900
1000
Inte
nsit
y [
nu
mb
er
of
co
un
ts]
GeTe 100 nm / SiO2 10 nm
2q [°]
120
130
140
150
160
170
180
190
200
Inte
nsit
y [
nu
mb
er
of
co
un
ts]
GeTe 100 nm / Ta 5 nm
GeTe 100 nm / TiN 5 nm
2q [°]
120
130
140
150
160
170
180
190
200
160
170
180
190
200
210
220
230
Inte
nsit
y [
nu
mb
er
of
co
un
ts]
2q [°]
Figure 4.8: Evolution of the GeTe 012 Bragg peak as a function of temperature observed for
Ψ=40 (heating rate of about 0.11C /min) for the 100nm thick GeTe films interfaced with
SiO2 , TiN and Ta. For each sample the thickest line in the graph corresponds to the first
temperature at which the Bragg peak becomes visible.
107
4.3 X-Ray Diffraction measurements
120 130 140 150 160 170 180 190 200 210 220 230
0.0
0.2
0.4
0.6
0.8
1.0
Cry
sta
llin
e f
ract
ion
Temperature [°C]
GeTe 100 nm / SiO2 10 nm
GeTe 100 nm / TiN 5 nm
GeTe 100 nm / Ta 5 nm
Figure 4.9: Evolution of the crystalline fraction as a function of temperature (heating rate
of about 0.11C /min) for the 100nm thick GeTe films interfaced with SiO2 , TiN and Ta as
obtained from the 012 GeTe Bragg peak area measured by XRD through in situ annealing for
a tilting angle Ψ=40.
In order to investigate a possible evolution of the samples textures with
temperature the area of the 012 Bragg peak has also been calculated as a function
of temperature for all the tilting angles. The results are shown in Fig. 4.10.
The crystallization temperature of each sample, calculated as the temperatures
corresponding to the midpoints of the rise steps of the plotted curves, remains
constant for all the values of Ψ and correspond to the values already reported
in Table 4.4. It can be noticed that at the beginning of crystallization the
relative peak intensities for different values of Ψ are the same as at complete
GeTe 100 nm Tx [C ]
SiO2 154
TiN 153
Ta 175
Table 4.4: Crystallization temperatures Tx of 100 nm thick GeTe films sandwiched between
TiN, Ta or SiO2 , obtained as the temperatures corresponding to the midpoints of the rise
steps of the Bragg peaks areas as a function of temperature reported in Fig. 4.9 (heating rate
of 0.11C /min).
108
4.3 X-Ray Diffraction measurements
100 120 140 160 180 200 220 240 260 280 300
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
100 120 140 160 180 200 220 240 260 280 300
-1000
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
100 120 140 160 180 200 220 240 260 280 300
-1000
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
GeTe 100 nm / SiO2 10 nm
Pe
ak
Are
a
Temperature [°C]
Psi = 0°
Psi = 20°
Psi = 40°
Psi = 60°
GeTe 100 nm / TiN 5 nm
Pe
ak
Are
a
Temperature [°C]
Psi = 0°
Psi = 20°
Psi = 40°
Psi = 60°
GeTe 100 nm / Ta 5 nm
Pe
ak
Are
a
Temperature [°C]
Psi = 0°
Psi = 20°
Psi = 40°
Psi = 60°
Figure 4.10: 012 Bragg peak area as a function of temperature (heating rate of 0.11C /min)
and of the sample tilting angle Ψ for 100 nm GeTe films sandwiched between TiN, Ta or SiO2 .
The points on the descending temperature ramp are also shown, and no evolution occurs during
the cooling down process.
109
4.4 Synchrotron X-Ray Diffraction
Ψ=0 [nm] Ψ=20 [nm] Ψ=40 [nm] Ψ=60 [nm]
SiO2 77 68
TiN 63 67 67 65
Ta 56 60 62 58
Table 4.5: Final mean grain sizes measured at 100C after annealing for different values of
the tilting angle Ψ for 100nm thick GeTe films.
crystallization only for the SiO2 interfaced sample. For both the TiN and Ta
interfaced samples, the most intense peaks at the beginning of crystallization
correspond to Ψ=40 and Ψ=60, while the most intense peak is found for
Ψ=20 at complete crystallization. This suggests an evolution of texture with
temperature.
The evolution of the grain sizes as a function of temperature, calculated us-
ing Sherrer analysis on the 012 Bragg peaks of all the samples and neglecting
any inhomogeneous strain (see appendix A.2.1), is shown in Fig.4.11. The in-
strumental resolution is FWHMinstr=0.295and the error bar for the grain size
is about ± 5 nm. The grain size has been calculated for each point of the in-
creasing temperature ramp and at 100C after annealing. For Ψ=0 no peaks
are detectable for the SiO2 interfaced samples and for Ψ=20 the signal is so
weak that no reliable results can be obtained, thus for Ψ=0 and Ψ=20 the
grain size has been calculated only for the TiN an Ta interfaced samples. When
possible, the grain size after in situ annealing has been calculated and is reported
in Table 4.5. For both Ψ=40 and 60, the sample interfaced with SiO2 has the
highest grain size, while the Ta interfaced sample has the smallest grain size and
intermediate values are found for the TiN interfaced sample. In particular, the
effect is more marled for Ψ=40. For both Ψ=0 and Ψ=20 the grain size is
slightly larger for the TiN interfaced sample than for the Ta interfaced one.
4.4 Synchrotron X-Ray Diffraction
A set of samples of 100 nm thick GeTe films interfaced with Ta, TiN or SiO2 has
been annealed at 400C for 15 minutes (equivalent heating rate around 26.7C /min)
110
4.4 Synchrotron X-Ray Diffraction
100 120 140 160 180 200 220 240 260 280 30030
35
40
45
50
55
60
65
70
100 120 140 160 180 200 220 240 260 280 30030
35
40
45
50
55
60
65
70
100 120 140 160 180 200 220 240 260 280 300
30
40
50
60
70
80
100 120 140 160 180 200 220 240 260 280 300
30
40
50
60
70
TiN
Ta
Gra
in s
ize
[n
m]
ψ=0°
Temperature [°C]
Temperature [°C]
TiN
Ta ψ=20°
SiO2
TiN
Ta ψ=40°
Gra
in s
ize
[n
m]
Temperature [°C]
SiO2
TiN
Ta
Gra
in s
ize
[n
m]
Gra
in s
ize
[n
m]
ψ=60°
Temperature [°C]
Figure 4.11: Grain size calculated by Sherrer analysis on the 012 Bragg peak of GeTe 100 nm
thin films interfaced with SiO2 , TiN or Ta. The error bar is about ± 5 nm. The dashed lines
indicate the final dimensions of grains, measured at 100C after annealing, and are reported
in Table 4.5.
111
4.5 Discussion and conclusions
and has been characterized by X-ray Diffraction on the BM02 CRG-D2AM
beamline (ESRF Grenoble, France) with an incident photon energy of 17.5 keV
photon (λ =0.707 A) and using a 2D CCD camera detector. A detailed descrip-
tion of the experimental setup, which is similar to the one used in Chapter 3,
can be found in appendix A.2.2. The 3D image of the SiO2 interfaced sample
shown in Fig. 4.12(a) clearly shows that the intensity of the partial diffracted
rings corresponding to GeTe 101, 012, 104 and 110 Bragg peaks is zero around
the center of the image and very high at the image borders, indicating a strong
texture. In Fig.4.12(b) the TiN interfaced sample is characterized by a light tex-
ture while the Ta interfaced sample in Fig.4.12(c) exhibit completely isotropic
rings. This result is in agreement with those obtained from the in situ XRD ex-
periments described in section 4.3 for the SiO2 interfaced sample. In particular,
the absence of signal measured at Ψ=0 agrees very well with what can be seen
in Fig. 4.12(a). However, the weak texture observed in section 4.3 for the Ta
sample is absent, while the texture of the TiN interfaced sample is stronger in
Fig.4.12(b) compared to what observed in section 4.3. This observation suggests
that the final texture is influenced by the anneling conditions. The samples were
identical in both experiments but at the synchrotron they have been annealed
at 400C for 15 minutes, while the samples of section 4.3 have been heated up
to 300C at a heating rate of 0.11C /min, staying at high temperature for a
long acquisition time.
4.5 Discussion and conclusions
An effect on the crystallization temperature of interfacing GeTe and GST films
with a Ta capping layer has been observed and confirmed both through XRD
and reflectivity measurements. First, the increment of Tx for a sample interfaced
with Ta compared to SiO2 or TiN interfaced ones exists even for films 100 nm
thick, indicating that it is not correlated to a size effect. Moreover, the activation
energy EA is higher for Ta interfaced samples than for the SiO2 interfaced ones.
The grain size measured for Ψ=40 is higher for the SiO2 interfaced sample
than for the TiN and Ta interfaced samples, and the SiO2 interfaced sample
112
4.5 Discussion and conclusions
Figure 4.12: 3D images of the diffracted rings obtained for GeTe 100 nm thin films annealed
at 400C for 15 minutes and interfaced with (a) SiO2 (b) TiN (c) Ta. The SiO2 interfaced
film is strongly textured while only a faint texture is visible for the TiN interfaced film and
the GeTe rings are isotropic for the Ta interfaced film. The strongly textured rings that can
be seen in (c) correspond to Ta.
113
4.5 Discussion and conclusions
SiO2 TiN Ta
GeTe 100 nm Tx [C ] 188 193 213
EA [eV] 2.58 not measured 3.83
Thickness dependence of Tx weak weak strong
Texture strong weak weak
Grain size for Ψ=40 [nm] 77 67 62
Table 4.6: Summary of the different characteristics observed for GeTe thin films encapsulated
in SiO2 , TiN or Ta. The SiO2 and Ta interfaced samples present the most relevant differences
in their properties, while the TiN interfaced sample can be considered as an intermediate
situation.
is strongly textured while it is not the case for the other samples. Finally, the
crystallization temperature remains almost constant for the SiO2 interface as the
PC films thickness is reduced from 100 nm to 10 nm, while it increases for the
the other samples, the effect being more marked on the Ta interfaced sample. All
these different characteristics are summarized in Table 4.6, where it is evident
that the samples interfaced with SiO2 and Ta exhibit relevant differences in their
properties while the TiN interfaced sample can be considered as an intermediate
situation. For this reason, the discussion will be focused mostly on the differences
between SiO2 and Ta samples.
In order to identify the preferred orientation in the SiO2 interfaced sample,
the diffracted intensity as a function of 2θ (already reported in Fig.4.6 for Ψ=0)
is shown in Fig.4.13 for each value of Ψ. The curves are vertically shifted for
clarity. It can be immediately noticed that while the 012 Bragg peak reaches
its maximum intensity for Ψ of about 40(see Fig.4.7), the 101, 202, 104 and
110 peaks exhibit the highest intensity for Ψ=20. It can be observed that the
angle between the (012) plane and the (010) plane is around 38, while the angle
between the (101) and (100) planes is around 21. This suggest that the grains
grow with a preferred orientation with the (100) or (010) plane parallel to the
sample surface.
It is worth noting that different values of EA and Tx could be explained by
an effect of doping due to the diffusion of the encapsulating material into the PC
114
4.5 Discussion and conclusions
24 28 32 36 40 44 48 52
0
500
1000
1500
2000
2500
Y =60°
Y =40°
Y =20°
Y =0°
20
2
02
10
06
11
3
01
5
11
0
10
4
10
1
00
3
01
2
Inte
nsi
ty [
nu
mb
er
of
cou
nts
]
2q [°]
Figure 4.13: Diffracted intensity as a function of 2θ for the GeTe 100 nm thin film interfaced
with SiO2 , measured at 100C after annealing at 300C (heating rate around 0.11C /min)
for various tilting angles Ψ. The intensity of the 012 Bragg peak is maximum at around
Ψ=40, as already reported in Fig.4.7, while the 101, 202, 104 and 110 peaks exhibit the
highest intensity for Ψ=20.
115
4.5 Discussion and conclusions
0 200 400 600 800 1000 1200
0
500
1000
1500
2000
2500
3000
3500
4000
0 200 400 600 800 1000 1200
0
1000
2000
3000
4000
GeTe 30 nm
Co
un
ts J
(1
/s)
Sputtering time A (s)
Ge+
Sb+
Te+
Ta+
GST 30 nm
Ge+
Sb+
Te+
Ta+
Co
un
ts J
(1
/s)
Sputtering time A (s)
Figure 4.14: Secondary Ion Mass Spectrometry (SIMS) measurements performed on GeTe
and GST 30 nm thin films sandwiched with Ta. By definition, the interface for each element
can be placed in the point at which half of the signal intensity is lost, corresponding to the
vertical lines in the figure. From those measurements the diffusion of Ta inside the GeTe and
GST layers is extremely low.
layer, as described in Chapter 2 in the case of N and C doping elements. In order
to exclude this effect, Secondary Ion Mass Spectrometry (SIMS) measurements
have been performed on GeTe and GST 30 nm thick films sandwiched in Ta.
The results, reported in Fig. 4.14, show a very weak diffusion of Ta inside GeTe
or GST films, thus establishing that the samples are not Ta-doped.
It is possible to find an explanation for the observed phenomena by making
the hypothesis that the SiO2 /GeTe interface is energetically more favorable for
a particular orientation than the Ta/GeTe interface. If this hypotesis is true,
it leads to some consequences that will be analyzed in the following. First, the
different grain dimensions obtained at Ψ=40for SiO2 and Ta interfaced samples
can be explained as a result of an abnormal grain growth along a preferred
direction [105]. If the interfacial energy between GeTe and SiO2 is minimized
for grain textured with the (012) plane tilted by around 40with respect to the
sample surface, then the the growth of grains with this orientation is preferred.
As a result, the grain sizes are larger for those grains and the sample is textured,
corresponding to what observed experimentally.
A second consequence of the hypothesis of an energetically favorable SiO2 interface
can be a different crystallization mechanism for the two samples. The crystal-
lization can begin as heterogeneous nucleation at the interfaces for the sample
116
4.5 Discussion and conclusions
sandwiched in SiO2 and then the growth of those nuclei is dominant in the
crystallization process, being more rapid than the homogeneous nucleation that
takes place in the bulk of the PC material. This situation in illustrated in Fig.
4.15 a. On the other hand, as represented in Fig. 4.15 b, for the sample inter-
faced with Ta the heterogeneous nucleation at the interfaces could be somehow
suppressed or slowed down, so that the crystallization process is now dominated
by the homogeneous nucleation which is known to be normally slower than the
heterogeneous nucleation (see section 1.3.3).
The difference in the crystallization mechanism between samples interfaced
with Ta and SiO2 is supported by the evolution of the crystallization tempera-
ture Tx with thickness. Considering the model of Zacharias reported in section
3.1.1, homogeneous nucleation can induce large variation of Tx with shrinking
thickness while it is not necessarily the case for heterogeneous nucleation. This
is coherent with what observed for Ta and SiO2 interfaced samples, as reported
in Table 4.6.
The situation of the sample interfaced with TiN is somehow in the middle
between the Ta and SiO2 interface cases. The value of Tx corresponds to the
one measured for the SiO2 interfaced sample, while the weak texture and the
grain size indicate no preferred orientation for grains growth. This suggests that
the crystallization begins with heterogeneous nucleation as in the SiO2 interfaced
sample case, but with no preferred grain orientation and no consequent abnormal
growth, leading to a weak texture of GeTe and a grain size comparable to the
one measured for the Ta interfaced sample.
Even if the assumption of an energetically favorable interface for SiO2 can
well explain the obtained results, some questions are still open. If the situation
depicted in Fig.4.15 is true, the SiO2 interface should promote heterogeneous nu-
cleation more than the Ta interface, which should inhibit the formation of nuclei
instead. This is unexpected because it looks unlikely that an amorphous could
trigger nucleation more easily than a metal. Indeed, in Ref.[12] the nucleation of
various Ge-Sb-Te compounds have been studied and no heterogeneous nucleation
at the PC material/substrate interface was observed, even for a SiO2 interface,
and the only nucleation observed is the one at the interface with the native oxide
117
4.5 Discussion and conclusions
Figure 4.15: Possible models of crystallization for (a) SiO2 and (b) Ta interfaced GeTe
thin films. Different colors correspond to different orientation of the grains. In the case of
SiO2 interface the crystallization begins with heterogeneous nucleation at the energetically
favorable interface and the nuclei grow with a preferred orientation before the homogeneous
nucleation starts. The final result are bigger grains with a preferred orientation. In the
case of Ta interfaced PC thin film, the heterogeneous nucleation at the interfaces is somehow
suppressed, thus the crystallization is driven by the homogeneous nucleation that starts later
respect the heterogeneous one, leading to a weak texture.
118
4.5 Discussion and conclusions
of the PC material. The homogeneous nucleation was not observed either. This
can be explained by considering that the native oxide for a Ge-Sb-Te material
is usually GeO or GeO2 so that the Ge-Sb-Te material is Ge-depleted at that
interface and this lowers Tx [6]. In the case of samples interfaced with a capping
material on both sides, as the films studied here, no native oxide should exist at
the interfaces so this preferential nucleation site is absent. However, considering
that the film of SiO2 is deposited by PVD, it can present an excess of Si or
O. In case of an O-rich SiO2 an oxidation at the GeTe/SiO2 interface cannot be
excluded, while in the case of a Si-rich SiO2 some nanocrystals of Si can possibly
form inside the oxide and act as nucleation sites [106].
Even if the effect of different interfaces on the Tx of a PC material is ev-
ident, the reasons behind this phenomena are still unclear. Further studies
are required, focused on confirming the hypothesis of an energetically favorable
grains orientation at the GeTe/SiO2 interface and the following consequences
on the nucleation and growth.Besides, studies by XPS experiments on the local
structure at the interfaces are required. TEM measurements could be useful in
order to investigate nucleation mechanism for different interfaces.
119
Conclusion
The present work gives a contribution to understand the properties of some
PC materials used in PCM devices and the effects of scaling and interface on
the amorphous to crystalline phase transformation.
The first part of the thesis has been dedicated to investigate the local struc-
ture of C and N doped amorphous GeTe. The aim was to understand the local
origin of better electrical properties of doped GeTe devices compared to un-
doped GeTe ones. The impact of doping was observed experimentally through
the appearance of a new peak in the pair distribution function of doped GeTe, in-
dicating the formation of a bond at a new distance that is absent in the undoped
amorphous material. The formation of new environments involving carbon and
nitrogen has been confirmed through ab-initio simulations. In the case of car-
bon doping, strong changes are induced at the second neighbor level through
tetrahedral and triangular units centered on carbon. The new peak observed
experimentally corresponds to new Ge-Ge distances in these units, while for N-
doping it correspond to the new Ge-Ge distance in tethahedral and pyramidal
units centered on nitrogen. It should be remarked that for C doped GeTe mea-
sured and calculated pair distribution functions are in good agreements, whereas
the agreement is not as good for N doped GeTe. One possible explanation, to
be confirmed, is that an important proportion of nitrogen form N2 molecules
in the film. Further steps require the study of the crystalline phase of doped
materials, in order to understand the role played by doping elements during
121
Conclusions
and after crystallization. This is extremely important from the point of view of
optimizing materials for a device.
The subject of the second part of this work has been the impact of con-
finement on GST crystallization. A fundamental requirement for further de-
velopment of PCM is their ability to be scaled without deterioration of their
properties. The results known from literature for thin films, reported in Chap-
ter 3, were not encouraging. In this thesis, nano-sized clusters of GST embedded
in Al2O3 have been made using a sputtering gas phase condensation source and
their crystallization has been studied through X-ray diffraction. The amorphous
to cubic crystalline phase transition has been unambiguously observed for clus-
ters. The crystalline clusters experience a tensile strain that can be ascribed to
the effect of the surrounding rigid Al2O3 matrix. The crystallization tempera-
ture of clusters is only slightly higher than that of a 10 nm thin film of GST
deposited under the same conditions. It is worth underlining that this result
is positive for PCM because it shows that the scaling effect on the crystalliza-
tion temperature in a phase change material can be small. Further studies are
required in order to clarify the origin of this difference in crystallization tempe-
rature, since many different effects can be involved. A composition effect (rising
from the fact that clusters are slightly Te-depleted compared to the thin film),
different surface to volume ratio, matrix influence, stress effects or an intrinsic
size effect could all play a role. The obtained results open new possibilities for
the study of nano-sized clusters of phase change materials. These particles have
been deposited by a method which is close to the one used for PCM thin films
deposition, thus giving information that can be easily exported to device fabri-
cation. It would be interesting to deposit smaller GST clusters, but with this
method their size distribution will be larger. Furthermore, it would be useful
to study if nanoclusters of other PC material, such as GeTe, show the same
evolution of Tx with scaling. Finally, a study of the variability of phase change
properties with size, i.e. identification of intrinsic vs extrinsic effects, should be
done. In addition, the electrical properties of clusters can be measured.
The third and last part of the thesis has been dedicated to the investiga-
tion of the interface material effect on the crystallization temperature of GeTe
122
Conclusions
and GST. It has been observed through reflectivity measurements that both
GeTe and GST interfaced with Ta show a crystallization temperature around
20C higher than the one of GeTe and GST interfaced with TiN or SiO2 . Even
if some studies in literature had evidenced the influence of interfaces over the
crystallization temperature of Ge-Sb-Te materials, such a remarkable difference
in Tx , due only to an interface effect, was never reported before. X-Ray diffrac-
tion results on GeTe showed that the SiO2 interfaced samples are characterized
by a strong texture while a weak texture is observed for Ta and TiN interfaced
samples. The results obtained for Tx and the samples texture can be explained
by supposing different nucleation and growth mechanisms for the different sam-
ples. Nucleation would begin at the SiO2 /GeTe interface and grains grow along
a preferred direction through the film thickness, originating a textured crys-
talline film. For Ta interfaced samples the nucleation at the Ta/GeTe interface
would be somehow suppressed and the nucleation would be only homogeneous
with no preferred orientation and no texture. The case of the TiN interface is
somehow intermediate. However, it is unclear how an amorphous interface can
promote nucleation of crystal phase more easily than a metallic interface. Ac-
tually, a simulation in COMSOL Multiphysics is already in progress in order to
quantify the crystallization parameters that are influenced by different interface
materials.
From a more general point of view, studies should move from a global to
a local approach. It will be interesting to check the nucleation mechanism for
different interfaces not only through simulation, but also through transmission
electron microscopy measurements. This technique will be useful to observe
the phase change in nanoclusters of Chapter 3. Similarly, X-ray photoelectron
spectroscopy can be used to study the chemical bonding near the interfaces and
to investigate the chemical bonding of C and N doped GST and GeTe samples,
both in the amorphous and crystalline phase.
Finally, let us recall that the goal of these studies is the understanding of
scaling and confining effect on advanced phase change materials used in PCM. In
this thesis, the effects of doping, scaling and interfaces have been addressed sep-
arately. The obtained results and the success of the cluster deposition method
123
Conclusions
open new perspectives. For instance, considering the remarkable interface effect
evidenced in thin films, it would be interesting to deposit and characterize dif-
ferent PC materials clusters embedded in different matrices as Ta, TiN, W or
SiO2 . Indeed, such nanoparticles offer a high surface to volume ratio that can
enhance the interface effect. Another perspective is to deposit N or C doped
clusters. This could allow to study the impact of doping on such small systems,
as well as the effect of doping concentration variability.
124
Appendix A
Experimental Techniques
125
A.1 Reflectivity measurements Conclusions
Figure A.1: Evidence of the different optical properties of the PC material Ge2Sb2Te5 in
the amorphous and crystalline phases. The reflectivity of GST is reported as a function of
temperature starting from an initially amorphous sample. The amorphous phase is character-
ized by a low reflectivity value compared to the one of the crystalline phase. On the graph it
is easy to identify the crystallization temperature at which the phase transformation occurs.
A.1 Reflectivity measurements
Reflectivity measurements consist in monitoring the reflectivity value of a PCM
sample as a function of temperature. The amorphous phase is characterized by
a lower reflectivity than the crystalline phase, so when the phase transforma-
tion occurs the measured reflectivity value increases. An example of a measured
curve has been shown in the Introduction of Chapter 1 (Figure 1.2). The crys-
tallization temperature can be defined as the point of maximum derivative of
the measured reflectivity curve (other definitions of Tx can be considered, as
discussed in section 3.1.1). The measured reflectivity can be normalized in or-
der to obtain the crystalline fraction.
A schematic description of the reflectometer used for the experiments per-
formed in this thesis is reported in Figure A.2. An amorphous sample is placed
in a vacuum chamber on a heating plate and the sample reflectivity is mea-
sured through a red laser beam (λ=670 nm). The plate temperature can reach
126
Conclusions A.2 X-Ray Diffraction
Figure A.2: Schematic representation of the reflectometer used for reflectivity measurements.
The laser beam is directed onto a birefringent filter and divided in two beams, and one of them
is directed to the sample. The direct and reflected beams are collected by a photodetector
and processed to obtain the measured signal.
a maximum value of 400C with a heating rate that can vary between 2C /min
and 20C /min. The temperature is continuously measured by a thermocouple
and controlled by a temperature feedback. The samples can be measured under
vacuum or in an argon atmosphere, in order to limit the oxidation at high tem-
peratures. It is important to remark that the crystallization can be observed
only for the part of sample in which the laser beam penetrates (around 30 nm).
The capping layer that is deposited over the phase change material should then
be sufficiently thin to let the beam reach the underlying layer. Moreover, the
effect of beam reflection by the capping layer can affect the measured reflectivity
and can even hide the PCM signal if the capping material is not transparent
enough to the laser beam.
A.2 X-Ray Diffraction
From the analysis of X-ray diffraction it is possible to deduce information on
the sample structure and microstructure.
When an incident wave interacts with the electrons of an atom, if no energy loss
occurs, the result is a new spherical wave with the same energy of the incident
127
A.2 X-Ray Diffraction Conclusions
1
2
Detector
SampleBeam
Beam
source
Figure A.3: Schematic representation of the geometry used for XRD analysis in laboratory.
The experiment is performed in a θ-2θ configuration (θ1 = θ2) and the tilting angle Ψ allows
to measure the sample texture.
wave (elastic scattering). When two or more atoms are involved, the resulting
spherical scattered waves interact by constructive or destructive interference
depending on their phases.
In this thesis, the XRD experiments have been performed by means of two
different X-Ray sources. The first one is a conventional source, an X-Ray tube
used in laboratory measurements. The second one is the synchrotron radiation
that is required for measuring very small systems (as the nanoclusters described
in Chapter 3) due to its high brilliance.
A.2.1 Conventional X-Ray Diffraction laboratory exper-
iment
In laboratory, the XRD experiments were performed in the θ-2θ configuration
with the geometry schematically reported in Fig.A.3. The variation of the tilting
angle Ψ allows to measure the sample texture. The instrument consists in a
XPERT PRO MRD diffractometer equipped with a Cu anode (λ=1.5406A) and
a PANalytical pixel point detector. A furnace (Anton Paar) is used for XRD
with in-situ annealing in the temperature range 100−300C . In order to get
the peak width and position the diffracted peaks are fitted by a Pseudo-Voigt
function. Supposing that there is no inhomogeneous strain effect, the grain size
can be calculated from the peak width by using the Scherrer’s formula
Dhkl =0.9 · λ
(Γmeas − Γinstr) cosθ(A.1)
128
Conclusions A.2 X-Ray Diffraction
where Dhkl is the grain size calculated for the direction perpendicular to the
〈hkl〉 plane, λ is the experimental wavelength, Γmeas is the measured Full Width
Half Maximum, Γinstr is the instrumental Full Width Half Maximum and 2θ is
the peak position. For the instrument used Γinstr = 0.295.
A.2.2 Large-scale facilities experiments
In this thesis, experiments have been performed at two different Synchrotron
beamlines, the CRISTAL beamline (SOLEIL, Saclay) for the experiments de-
scribed in Chapter 2 and the BM02 CRG-D2AM beamline (ESRF, Grenoble)
for the experiments reported in Chapters 3 and 4 respectively. In both cases
the intensity as a function of 2θ is obtained by integration over a bidimensional
image but the experimental methods differ. On beamline CRISTAL X-ray scat-
tering has been measured up to large scattering angles in order to determine
the pair distribution function, while at ESRF selected Bragg peaks have been
studied.
SOLEIL data analysis
The experiment at the synchrotron SOLEIL have been performed in transmission
geometry over the powder samples described in Chapter 2. The experimental
setup is shown in Fig.A.4. The incident beam has an energy of of E=45.4793
keV (λ = 0.4441A) and the transmitted scattered beam is collected by an image
plate detector MAR350 (3450x3450 100µm pixels). An example of image is re-
ported in Fig.A.5 for an amorphous GeTe sample. The distance D between the
sample and the image plate was equal to ≈ 21cm and could not be reduced (see
Fig.A.4). Considering that obtaining a good quality structure requires measure-
ments in a large range of Q, and considering that the distance D could not be
reduced further, one chose a configuration where the center of the image plate
does not coincide with the incident beam as can be seen in Fig.A.5. This allows
to obtain a wider range of Q (up to 20.1 A−1) by integrating over a vertical
sector. One drawback of this set-up is a loss of intensity since only a fraction of
the diffraction ring is selected for integration.
129
A.2 X-Ray Diffraction Conclusions
Figure A.4: Picture and schematic representation of the experimental setup used at the
synchrotron SOLEIL. The scattered transmitted beam is collected by an image plate detector
placed at a distance D≈ 21cm from the sample, which is the minimum allowed distance in
this configuration. Thus, in order to obtain a high value of Q, the center of the image does
not correspond to the center of the detector.
130
Conclusions A.2 X-Ray Diffraction
The counting time for one image was 300 s. Each image has been corrected by
subtracting the dark image acquired without beam in the same conditions. All
the images have been treated by using the freeware software Fit2D [107]. It is
necessary to first correct the image for the geometrical errors due to the tilting
of the image plate with respect to the plane perpendicular to the incident beam.
The tilt parameters as well the distance between the sample and the detector
can be obtained by analysing the image of a LaB6 powder sample measured in
the same conditions. Each image can then be integrated. The integration sector
is shown in Fig.A.5. The integration process takes into account polarization
corrections. The value of polarization was 0.96.
The resulting intensity as a function of 2θ must finally be multiplied by 1−e(−dµIPcos2θ )
1−e−dµIP
in order to take into account the incomplete absorption of photons in the active
layer of the image plate of thickness d and absorption coefficient µIP .
In order to extract the scattering from the sample from the intensity measured
on the filled capillary, it is necessary to substract the scattering by the empty
capillary, taking into account the sample transmission, and to remove the scat-
tering by air. For each case (filled capillary, empty capillary or air) 4 images
have been acquired for 300 s and the obtained integrated intensities summed up.
The integrated intensity scattered by the sample I(2θ) is obtained by using the
following relation
I(2θ) = [(Iraw − Iair)− t · (Iemptycap − Iair)] ·1− e(−
µIPcos2θ )
1− e−µIP(A.2)
where Iraw is the raw intensity measured on the filled capillary , Iair is the air
signal, Iemptycap is the intensity of the empty capillary and t is the measured
transmission coefficient. The obtained quantity has then been processed by
using the PDFgetX2 software [108] which removes the Compton and fluorescence
signals, converts 2θ in Q and by adequate normalization gives the structure factor
S(Q). It should be emphasized that all the data corrections and normalization
must be performed carefully since any inadequacy directly affects the asymptotic
behaviour of S(Q). If S(Q) is not tending to 1 at large Q the pair distribution
function obtained by Fourier transform of Q(S(Q)-1) (see Eq. 2.18 in Chapter
2) is dramatically affected.
131
A.2 X-Ray Diffraction Conclusions
Figure A.5: Image acquired on the capillary containing amorphous GeTe. Only the pixels
contained in the vertical triangular sector shown on the figure have been selected for integra-
tion. The pixels that appear as a white spot at the center of the image plate are damaged
and must be avoided.
132
Conclusions A.2 X-Ray Diffraction
Figure A.6: Schematic representation of the experimental setup used at the synchrotron
ESRF. The diffracted beam is collected by a CCD camera at a distance D from the sample
of around 20 cm. The camera and the sample are tilted of an angle θ2 and θ1 with respect
to the incident beam, respectively. If θ1 = θ2, the configuration is in a strict θ −2θ geometry
only at the center of the camera.
ESRF data analysis
The diffraction experiments at the ESRF synchrotron (beamline BM02 CRG-
D2AM) have been performed in a θ-2θ configuration or in pseudo-grazing inci-
dence, over thin film samples and clusters samples described in Chapters 4 and
3.
The experimental setup is shown in Fig.A.6. The incident beam is directed
onto the sample, and the diffracted beam is collected by a CCD camera placed at
a distance D from the optical center. The CCD camera is a screen of 1340x1300
squared pixels of 50 µm. The angle between the camera and the incident beam
(θ2) has been chosen to be 17.5for all the performed experiments, while the an-
gle between the incident beam and the sample surface (θ1) was 8.75for the θ-2θ
configuration and 4for the pseudo-grazing configuration used in Chapter 3. The
incident angle of 4cannot be strictly considered as a grazing incidence, but for
smaller angles, the beam spot over the sample would have been larger than the
133
A.2 X-Ray Diffraction Conclusions
sample size (between 25nm2 and 400nm2). The diffracted beams are collected in
an angular range ±α in the horizontal direction and ±β in the vertical direction.
Strictly speaking, for θ1=8.75the geometry is in a real θ-2θ configuration only
at the center of the image. For all other points, the measured intensity collected
by the CCD camera is not due to diffraction of planes parallel to the sample
surface. In particular, at point B of Fig.A.6 is collected the beam diffracted
by a plane rotated by β/2 around an axis perpendicular to the figure plane.
Considering that the CCD camera is almost squared, α ≈ β and their values
depend on the distance D that determines the angular aperture measured by the
camera. For example, for the experiment on cluster described in Chapter 3 the
signal is measured in a 2θ range from 8 to 26, meaning that α ≈ β ≈ 9. The
exact value of the distance D is calculated for each experiment by measuring the
position of the direct beam on the camera without sample for θ2 values varying
between -6and 6.
For the experiments described in Chapter 3 on clusters and thin films, the in-
tensity as a function of 2θ has been obtained from each bidimensional image by
integrating over the rings (excluding only a border of 100 pixels on each side of
the image). A dedicated program, developed on the BM02 beamline, has been
used for integration. This program takes into account geometrical corrections,
in particular the fact that the camera plane is not vertical so that the diffrac-
tion rings are not circles. The resulting angular integrated intensities includes
the contribution of all the planes contributing to the measured partial rings on
the image, so any information on the texture of samples between this angular
aperture is lost in the final result. This means that the final integrated peak po-
sitions and widths are the sum of the peaks over the various Ψ, so care must be
taken in analyzing their width [95]. However, for non textured samples, as the
clusters characterized in Chapter 3, the rings are isotropic and no information
is lost with integration.
The samples described in Chapter 3 (GST thin film and nanoclusters) have
been studied by XRD as deposited, after ex-situ annealing and also during in
situ annealing experiment. In all cases, in order to get the signal of the studied
material it is necessary to subtract the contribution of the Si substrate and of
134
Conclusions A.2 X-Ray Diffraction
the Al2O3 encapsulating material from the measured signal. For this purpose,
the signal of a blank sample constituted of a 20 nm thick Al2O3 film deposited
on a Si substrate has been measured. However, the intensity of the substrate
and Al2O3 signal in sample measurement is not exactly the same as the one
measured for the blank sample, meaning that the intensity diffracted by GST,
IGST , should be obtained through the relation
IGST = Imeasured − αISi+Al2O3
(A.3)
where Imeasured is the intensity measured for the sample thus including the sub-
strate and Al2O3 contributions, ISi+Al2O3
is the intensity measured for the
blank sample and α is a coefficient that needs to be adjusted for each sample.
The fact that α is not equal to 1 can be due to many reasons, including a dif-
ference in the quantity of Al2O3 matter between the samples. It is important to
determine α accurately since ISi+Al2O3
is much larger than IGST . However, a
perfect subtraction of the substrate signal could not be achieved. This is prob-
ably due to the fact that the relative contributions of the Al2O3 signal and Si
signal are not the same in the blank sample and in presence of GST.
For the in situ experiments of Chapter 3, the furnace used is the same Anton
Paar furnace as for laboratory experiments (section A.2.1), equipped with a
PEEK dome. This dome gives an extremely intense undesired diffracted signal,
which is two orders of magnitude higher than the GST film signal and even three
orders of magnitude higher than the GST clusters signal.
135
Appendix B
Deposition method
137
Deposition method Conclusions
All the thin film studied in this thesis have been deposited by Physical Vapor
Deposition (PVD) through sputtering. In this method, a thin film is deposited
on the surface of a substrate by condensation of a vapor. The vapor is obtained
by bombarding a solid target (cathode) with a plasma and the ejected material is
directed onto the substrate (anode). In order to confine plasma on the surface of
the target, sputtering sources are equipped with magnetrons that utilize strong
electric and magnetic fields to increase the number of ionizing collisions near the
target surface.
The deposition apparatus is the Equipement Alliance Concept Cluster ACT200
(BHT, CEA Leti), which is composed of three deposition chambers. The co-
sputtering chamber is equipped with three targets of 100 mm of diameter that
can be used independently from each other or in co-sputtering to deposit different
compounds, and the substrate is a 200 mm Si wafer. The film uniformity is
improved by a satellite and sweeping motion of the target. The deposition is
done under Ar atmosphere at a pressure of 5 10−3 bar. The two other deposition
chambers are used to deposit electrodes and capping materials such as Ti, TiN
or Ta, while all the phase change thin films are deposited in the co-pulverization
chamber. The C-doped GeTe samples studied in Chapter 2 have been deposited
into the co-sputtering chamber by using two targets, one of GeTe and another
one of C. The N-doped GeTe samples have been deposited by pulverization of
a GeTe target in an Ar+N2 atmosphere. The film thickness is controlled by the
deposition time and the power applied to the targets. The thickness uniformity
is checked after deposition by 9 points measurements along the wafer diameter
and the films compositions are checked through Rutherford Back Scattering
(RBS).
138
List of Figures
1.1 Basic principle of the phase transformation [8]. PC materials can switch
reversibly between an amorphous state, corresponding to the logical level ’0’
or RESET, and a crystalline state corresponding to the logical level ’1’ or
SET. The SET operation consists in programming the cell into the SET state,
while the RESET operation consists in programming the cell into the RESET
state. To obtain the amorphous phase the PC material must be annealed
above its melting temperature and then rapidly cooled down. To obtain
the crystalline phase the material must be annealed above its crystallization
temperature Tx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 Evidence of the different optical properties of the PC material Ge2Sb2Te5 in
the amorphous and crystalline phases. The reflectivity of GST is reported
as a function of temperature starting from an initially amorphous sample.
The amorphous phase is characterized by a low reflectivity value compared
to the one of the crystalline phase. On the graph it is easy to identify the
crystallization temperature at which the phase transformation occurs. . . . 14
1.3 Schematic representation of the lance-like structure of a PCM cell device.
The PC material is interfaced with a top electrode and a bottom electrode
(heater). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 Current pulses for the programming operation of the cell. RESET pulse (a)
SETMIN pulse (b) and SET pulse (c). . . . . . . . . . . . . . . . . . 16
1.5 I-V characteristic of a PCM cell in the crystalline and amorphous states (from
Ref.[9]). The I-V characteristic of the amorphous state present a snap-back
in correspondence of a threshold voltage that is not present in the crystalline
I-V curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
139
LIST OF FIGURES LIST OF FIGURES
1.6 Time-temperature-transformation (TTT) diagram for a PC material taken
from Reference [8]. The phase transformation of a fixed volume of PC mate-
rial is reported depending on the time spent at a certain temperature. The
two orange lines on the graph indicate two different constant rate quenching
processes while the two purple lines indicate two annealing processes starting
at room temperature. . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Evolution of ∆Gcluster(r) as a function of r corresponding to Eq. 1.2, taken
from Chapter 7 of Reference [4]. The curve exhibit a maximum for the r = rc
(critical radius) that corresponds to the critical work for cluster formation
∆Gc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.8 Model for the heterogeneous nucleation taken from Chapter 7 of Reference
[4]. The crystalline cluster is a spherical cap which correspond to the ex-
posed part of a sphere of radius r. In the schematic picture are also re-
ported the wetting angle θ and the crystal-substrate, amorphous-substrate
and amorphous-crystal interfacial energies (respectively σcs, σls and σlc). . 24
1.9 PC materials reported on the ternary Ge:Sb:Te phase diagram, with the
GeTe− Sb2Te3 pseudo-binary line put in evidence (taken from Ref. [8]). . 28
1.10 Structure of GST in its crystalline metastable phase. One sublattice is oc-
cupied by Te atoms (light blue) while the other is randomly occupied by
Ge or Sb atoms (dark blue) or vacancies (around 20% ). The cubic lattice
parameter is 6.03 A [26]. . . . . . . . . . . . . . . . . . . . . . . . . 30
1.11 Structure of crystalline GeTe in its rhombohedral phase. The structure can
be described as a rocksalt-like structure, distorted by a relative shift of the
sublattices along the [111] direction. It is characterized by long (3.127 A)
and short (2.87 A) Ge-Te bonds shown respectively in white and green. . . 31
2.1 Low field cell resistance as a function of progam current for GST and GeTe
cells for various programming pulse times [24]. It can be noted that the SET
operation for the GeTe cell is faster and the difference in the resistance of
the amorphous and crystalline phases is higher. . . . . . . . . . . . . . 37
140
LIST OF FIGURES LIST OF FIGURES
2.2 Calculation of the activation energy EA by interpolation of the fail times as
a function of 1/kT . In order to obtain the fail time, a PCM cell is written
in the RESET state and the fail time is defined as the time at which the
resistance of the cell is reduced by one half. . . . . . . . . . . . . . . . 38
2.3 Reflectivity measurements of C and N doped GeTe films (150 nm thick)
[45]. In both cases Tx increases with increasing doping concentration and
the effect is stronger for C doping. . . . . . . . . . . . . . . . . . . . 39
2.4 Activation energy (left) calculated for undoped and C-doped GeTe and low
electric field resistance as a function of the programming current (right) for
a GST, undoped GeTe and C-doped GeTe cell [44]. The activation energy
increases and the RESET current decreases with doping. . . . . . . . . . 39
2.5 Schematic representation of an X-ray incident beam scattered by a point-like
sample. The incident wavevector is k0, the scattered wavevector is kf and
the momentum transfer is Q = k0 − kf . . . . . . . . . . . . . . . . . . 41
2.6 Measured (a) S (Q) and (b) g (r) for undoped amorphous GeTe. It can be
noted that S (Q) tends to 1 for high Q. . . . . . . . . . . . . . . . . . 48
2.7 Comparison between the measured g(r) of amorphous (blue) and crystalline
(red) undoped GeTe. . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.8 Measured g (r) for (a) undoped GeTe and GeTeC (C=9.6% and 16.3%) and
(b) undoped GeTe and GeTeN (N=4% and 10%). In both cases, the first peak
is constant with doping while the intensity of the second peak of the undoped
sample decreases with increasing doping contents. A new peak appears at
around 3.5 A in the doped samples. These effects increase as a function of
doping and are stronger in the GeTeN case. . . . . . . . . . . . . . . . 50
2.9 Comparison between measured and calculated g (r) for undoped GeTe, GeTeC
(C=16.3% in the experiment and C=15% in the simulation) and GeTeN
(N=10% both in the experiment and in the simulation). Even if an already
known shift between peaks positions can be observed, the evolution of the
simulated and measured pair distribution functions with doping are in good
agreement. The effect of N-doping is stronger in the calculated g (r) than in
the measured one. . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
141
LIST OF FIGURES LIST OF FIGURES
2.10 Partial pair distribution functions for (a) Ge-Ge, Te-Te, and Ge-Te pairs in
doped and undoped samples and (b) pairs involving C or N. Curves are shifted
for clarity. A new peak appears in the range 3.1-3.5 A in the Ge-Ge partial
pair distribution function for both C doped and N doped samples, while it
is absent in the undoped sample. A difference between partial contributions
involving C and the ones invloving N is the absence of Te-N bonds at small
distances (less than 3.2 A). . . . . . . . . . . . . . . . . . . . . . . . 53
2.11 Snapshot of the final state of the simulation box for GeTeC. Ge atoms are
represented in pink, Te atoms in light blue and C atoms in red. The in-
spection of this box combined with a bond angle analysis around C atoms,
reveals the presence of a mixture of tetrahedral (C − TeGe3, C − Ge4 and
C − Ge2Te2), triangular (C − C − Ge2 and C − C − GeTe), and linear (C
chains) bonds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.12 Summary of the carbon environments in the C-doped GeTe sample. C −
TeGe3, C−Ge4 and C−Ge2Te2 tetrahedra can be found, as well as C−C−Ge2
and C− C−GeTe triangular environments. . . . . . . . . . . . . . . . 56
2.13 Snapshot of the final state of the simulation box for GeTeN. Ge atoms are
represented in pink, Te atoms in light blue and N atoms in green. N − Ge3
pyramidal environments, N−Ge4 tetrahedral environments and N2 molecules
can be found. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.14 Summary of the nitrogen environments found in the N-doped GeTe sample.
N−Ge4 tetrahedra, N−Ge3 pyramids and N2 molecules have been observed. 58
3.1 Resistivity as a function of time at room temperature for thin films of GST of
different thicknesses pre-annealed at 143.5C (from Ref.[67]). The incubation
time τ , defined as the time elapsed before the onset of crystallization, and the
transition time from the highest to lowest resistivity increase with decreasing
film thickness, meaning that the crystallization speed is reduced for small
thicknesses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2 Model used Ref.[71] to interpret the thickness-dependent variation of Tx .
A cylindric crystalline nucleus is embedded in the amorphous phase, sand-
wiched between two oxide interfaces. . . . . . . . . . . . . . . . . . . 66
142
LIST OF FIGURES LIST OF FIGURES
3.3 Crystallization temperature Tx as a function of film thickness for various PC
materials: GST, N-doped GST (NGST), Ge15Sb85 (GeSb), Sb2Te and Ag-
and In-doped Sb2Te (AIST) deposited on Si and capped with Al2O3 , fitted
using Eq.3.1, as presented in Ref.[70]. . . . . . . . . . . . . . . . . . . 69
3.4 Sheet resistance of multilayered films of GST/SiO2 as a function of annealing
temperature, as reported in Ref.[78]. The label M25, M10 and M5 indicate
different bilayer thicknesses (M5 corresponds to the thinnest sample, M25 the
thickest one). The dotted lines correspond to ex situ annealing temperatures
used for further analysis in Ref.[78]. . . . . . . . . . . . . . . . . . . . 72
3.5 Scanning Electron Microscopy (SEM) image of as-grown GST nanowires from
Ref.[81]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.6 Measured values for the (a) recrystallization time at fixed temperature, (b)
nucleation rate and (c) activation energy as a function of nanowires diameter
as reported in Ref.[82]. . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.7 Schematic drawing of the apparatus used to deposit the nanocluster sam-
ples as long as the thin film samples used for comparison. The deposition
procedure is briefly illustrated. . . . . . . . . . . . . . . . . . . . . . 78
3.8 (a) TOF size distribution which shows that the nanoclusters have an average
diameter of 5.7 nm with a narrow size distribution (± 1 nm at half maxi-
mum) (b) TEM images of GST as-deposited clusters which indicate that the
particles are spherical and amorphous. These images have been made on
a low density dedicated sample and the clusters state and shape have been
checked over clusters with the largest diameter. (By courtesy of M. Audier,
LMGP CNRS, Grenoble INP-Minatec . . . . . . . . . . . . . . . . . . 79
3.9 Scanning Electron Microscopy (SEM) image of GST deposited clusters. The
red circle has a diameter of 20 nm. No trace of particles coalescence can be
seen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.10 (a) GST clusters sample: 4 layers of clusters capped with Al2O3 are deposited
on an Al2O3 substrate and capped again with Al2O3 . The average distance
between clusters in a layer is about 2 cluster diameters. (b) GST film sample:
10 nm thick film of GST sandwiched between two 10 nm thin Al2O3 films.
Both the clusters and film samples are deposited on a substrate of Si. . . . 81
143
LIST OF FIGURES LIST OF FIGURES
3.11 (a) X-ray 2D diffraction images for 200C ex situ annealed GST thin film.
(b) Same measurements for 200C ex situ annealed clusters. In both cases,
the 2D image of the blank sample (Si+Al2O3 ) has been subtracted. . . . 83
3.12 X-ray diffraction spectra at room temperature for as-deposited and 200C ex
situ annealed GST film, after background subtraction, with curves shifted for
clarity. Arrows indicate bulk GST fcc peak positions calculated assuming the
lattice parameter of GST a=6.0117 A reported in Ref.[25]. . . . . . . . . 84
3.13 X-ray diffraction spectra at room temperature for as-deposited and 200C ex
situ annealed GST clusters after background subtraction. Curves are shifted
for clarity. Arrows indicate bulk GST fcc peak positions calculated assuming
the lattice parameter of GST a=6.0117 A reported in Ref.[25]. . . . . . . 85
3.14 (220) diffraction peak for in situ annealed GST thin film at different temper-
atures. The dotted line is the peak, measured at room temperature, of the
thin film annealed ex situ at 200C (shifted for clarity). The arrow indicates
the calculated bulk GST peak position. . . . . . . . . . . . . . . . . . 87
3.15 X-ray diffraction spectra for in situ annealed GST clusters at different tem-
peratures. (a) (200) diffraction peak and (b) (220) diffraction peak. Curves
are evenly shifted to ease viewing. Dotted lines indicate the peak position,
measured at room temperature, of the 200C ex situ annealed clusters. . . 88
3.16 Normalized integrated intensities for (220) and (200) diffraction peaks for
GST clusters and for GST film as a function of temperature. The normalized
integrated intensities have been obtained through the relation Inorm = I/Imax
where I is the measured integrated intensity at a given temperature and
Imax is the maximum value of the integrated intensity (which correspond
to complete crystallization). The dotted lines indicate the crystallization
temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
144
LIST OF FIGURES LIST OF FIGURES
3.17 The Al2O3 matrix surrounding the cluster forces its volume to remain equal
to the one of the amorphous phase even after crystallization. The cluster in its
amorphous phase occupies a certain volume (a). When crystallization occurs,
the cluster volume tends to reduce of around 5% (b), but the embedding
Al2O3 matrix exerts a tensile strain over the cluster (c) thus forcing the
cluster to keep the volume corresponding to the amorphous phase, with the
effect of increasing the lattice parameter of the crystalline cluster (d). . . . 91
4.1 Structure of PC material thin films samples used in order to investigate the
interface effect on crystallization. The PC material can be either GeTe or
GST, of various thicknesses, sandwiched between (a) SiO2 , (b) TiN or (c)
Ta. All the samples have been deposited by sputtering as described in B. . 97
4.2 Crystalline fraction as a function of temperature obtained from reflectivity
measurements for (a) GeTe and (b) GST 100 nm thin films sandwiched be-
tween TiN, Ta or SiO2 heated at 10C /min. For both GeTe and GST thin
films the amorphous to crystalline transition occurs at a higher temperature
when the film is sandwiched between Ta. . . . . . . . . . . . . . . . . 98
4.3 Crystalline fraction as a function of temperature from reflectivity measure-
ments for (a) GeTe 30 nm and (b) GeTe 10 nm thin films sandwiched between
TiN, Ta or SiO2 , heated at 10C /min. For 30 nm thick GeTe films the amor-
phous to crystalline phase transition occurs clearly at a higher temperature
when the PC material is sandwiched between Ta. For 10 nm thick films of
GeTe the measurement becomes difficult for samples interfaced with TiN and
Ta, while Tx can be still easily identified for the SiO2 interfaced sample. . 101
4.4 Kissinger plot for GST 100 nm thin films sandwiched between Ta or SiO2 .
The absolute value of the line slope corresponds to the activation energy EA .
The points in graph have been calculated from the Tx value obtained for four
different heating rates r as reported in Table 4.3 . . . . . . . . . . . . . 102
4.5 Schematic representation of the experimental geometry of the XRD experi-
ment, where θ is the incident beam angle and Ψ is the tilting angle of the
sample. A detailed description of the experimental setup is provided in ap-
pendix A.2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
145
LIST OF FIGURES LIST OF FIGURES
4.6 Diffracted intensity as a function of 2θ for Ψ =0of GeTe 100 nm thin films
sandwiched between TiN, Ta or SiO2 measured at 100C after annealing at
300C with a heating rate of about 0.11C /min. The vertical lines cor-
respond to the calculated position of Bragg peaks for rhombohedral GeTe
(hexagonal indexation) [25]. No difference in the diffraction spectra can be
observed between the Ta and TiN interfaced samples, while no peaks are
visible for the sample sandwiched in SiO2 . The intense peak around 33 is
due to Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.7 012 Bragg peak for 100 nm thick GeTe films interfaced with SiO2 , TiN
and Ta measured at 100C after annealing at 230C (heating rate of about
0.11C /min) for various tilting angles Ψ of the samples. The Ta and TiN in-
terfaced samples show a weak texture while the sample sandwiched in SiO2 is
strongly textured, with a maximum peak intensity for Ψ=40 and no inten-
sity for Ψ=0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.8 Evolution of the GeTe 012 Bragg peak as a function of temperature observed
for Ψ=40 (heating rate of about 0.11C /min) for the 100nm thick GeTe
films interfaced with SiO2 , TiN and Ta. For each sample the thickest line
in the graph corresponds to the first temperature at which the Bragg peak
becomes visible. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.9 Evolution of the crystalline fraction as a function of temperature (heating
rate of about 0.11C /min) for the 100nm thick GeTe films interfaced with
SiO2 , TiN and Ta as obtained from the 012 GeTe Bragg peak area measured
by XRD through in situ annealing for a tilting angle Ψ=40. . . . . . . . 108
4.10 012 Bragg peak area as a function of temperature (heating rate of 0.11C /min)
and of the sample tilting angle Ψ for 100 nm GeTe films sandwiched between
TiN, Ta or SiO2 . The points on the descending temperature ramp are also
shown, and no evolution occurs during the cooling down process. . . . . . 109
4.11 Grain size calculated by Sherrer analysis on the 012 Bragg peak of GeTe 100
nm thin films interfaced with SiO2 , TiN or Ta. The error bar is about ±
5 nm. The dashed lines indicate the final dimensions of grains, measured at
100C after annealing, and are reported in Table 4.5. . . . . . . . . . . . 111
146
LIST OF FIGURES LIST OF FIGURES
4.12 3D images of the diffracted rings obtained for GeTe 100 nm thin films an-
nealed at 400C for 15 minutes and interfaced with (a) SiO2 (b) TiN (c)
Ta. The SiO2 interfaced film is strongly textured while only a faint texture
is visible for the TiN interfaced film and the GeTe rings are isotropic for
the Ta interfaced film. The strongly textured rings that can be seen in (c)
correspond to Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.13 Diffracted intensity as a function of 2θ for the GeTe 100 nm thin film inter-
faced with SiO2 , measured at 100C after annealing at 300C (heating rate
around 0.11C /min) for various tilting angles Ψ. The intensity of the 012
Bragg peak is maximum at around Ψ=40, as already reported in Fig.4.7,
while the 101, 202, 104 and 110 peaks exhibit the highest intensity for Ψ=20.115
4.14 Secondary Ion Mass Spectrometry (SIMS) measurements performed on GeTe
and GST 30 nm thin films sandwiched with Ta. By definition, the interface
for each element can be placed in the point at which half of the signal in-
tensity is lost, corresponding to the vertical lines in the figure. From those
measurements the diffusion of Ta inside the GeTe and GST layers is extremely
low. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.15 Possible models of crystallization for (a) SiO2 and (b) Ta interfaced GeTe
thin films. Different colors correspond to different orientation of the grains.
In the case of SiO2 interface the crystallization begins with heterogeneous
nucleation at the energetically favorable interface and the nuclei grow with
a preferred orientation before the homogeneous nucleation starts. The final
result are bigger grains with a preferred orientation. In the case of Ta inter-
faced PC thin film, the heterogeneous nucleation at the interfaces is somehow
suppressed, thus the crystallization is driven by the homogeneous nucleation
that starts later respect the heterogeneous one, leading to a weak texture. . 118
147
LIST OF FIGURES LIST OF FIGURES
A.1 Evidence of the different optical properties of the PC material Ge2Sb2Te5 in
the amorphous and crystalline phases. The reflectivity of GST is reported
as a function of temperature starting from an initially amorphous sample.
The amorphous phase is characterized by a low reflectivity value compared
to the one of the crystalline phase. On the graph it is easy to identify the
crystallization temperature at which the phase transformation occurs. . . . 126
A.2 Schematic representation of the reflectometer used for reflectivity measure-
ments. The laser beam is directed onto a birefringent filter and divided in
two beams, and one of them is directed to the sample. The direct and re-
flected beams are collected by a photodetector and processed to obtain the
measured signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
A.3 Schematic representation of the geometry used for XRD analysis in labora-
tory. The experiment is performed in a θ-2θ configuration (θ1 = θ2) and the
tilting angle Ψ allows to measure the sample texture. . . . . . . . . . . . 128
A.4 Picture and schematic representation of the experimental setup used at the
synchrotron SOLEIL. The scattered transmitted beam is collected by an
image plate detector placed at a distance D≈ 21cm from the sample, which
is the minimum allowed distance in this configuration. Thus, in order to
obtain a high value of Q, the center of the image does not correspond to the
center of the detector. . . . . . . . . . . . . . . . . . . . . . . . . . 130
A.5 Image acquired on the capillary containing amorphous GeTe. Only the pixels
contained in the vertical triangular sector shown on the figure have been
selected for integration. The pixels that appear as a white spot at the center
of the image plate are damaged and must be avoided. . . . . . . . . . . 132
A.6 Schematic representation of the experimental setup used at the synchrotron
ESRF. The diffracted beam is collected by a CCD camera at a distance D
from the sample of around 20 cm. The camera and the sample are tilted of
an angle θ2 and θ1 with respect to the incident beam, respectively. If θ1 = θ2,
the configuration is in a strict θ −2θ geometry only at the center of the camera.133
148
List of Tables
1.1 Properties that characterize PC materials [7]. . . . . . . . . . . . . . . 27
1.2 Comparison between the main properties of Ge2Sb2Te5 (GST) and GeTe. . 29
2.1 Crystallization temperatures Tx of C and N doped GeTe films (150 nm thick),
taken as the midpoint of the rising steps of the reflectivity curves reported
in Fig.2.3 [45]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Measured mass densities and atomic number densities for Ge52Te48, undoped
and doped with carbon or nitrogen, expressed in g/cm3 and atoms/A3, re-
spectively. The mass densities have been measured by X-ray reflectivity (XRR). 46
3.1 Crystallization temperatures as a function of heating rate and GST film
thickness from Ref.[67]. . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Peak positions and relative intensities for the fcc GST phase as expected from
a powder pattern. They have been estimated in a θ −2θ geometry at the
actual experimental wavelength considering the lattice parameter a=6.0117
A reported in Ref [25]. . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.1 Crystallization temperatures Tx of GeTe and GST 100 nm thick films sand-
wiched between TiN, Ta or SiO2 as obtained from the reflectivity measure-
ments of Fig. 4.2 for a heating rate of 10C /min. . . . . . . . . . . . . 99
4.2 Crystallization temperatures Tx of GeTe films 30 nm and 10 nm thick sand-
wiched between TiN, Ta or SiO2 as obtained from the reflectivity measure-
ments of Fig. 4.3 for a heating rate of 10C /min. . . . . . . . . . . . . 100
4.3 Crystallization temperature Tx obtained from reflectivity measurements for
different heating rates for GeTe 100 nm thin films sandwiched between Ta or
SiO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
149
LIST OF TABLES LIST OF TABLES
4.4 Crystallization temperatures Tx of 100 nm thick GeTe films sandwiched be-
tween TiN, Ta or SiO2 , obtained as the temperatures corresponding to the
midpoints of the rise steps of the Bragg peaks areas as a function of tempe-
rature reported in Fig. 4.9 (heating rate of 0.11C /min). . . . . . . . . 108
4.5 Final mean grain sizes measured at 100C after annealing for different values
of the tilting angle Ψ for 100nm thick GeTe films. . . . . . . . . . . . . 110
4.6 Summary of the different characteristics observed for GeTe thin films encap-
sulated in SiO2 , TiN or Ta. The SiO2 and Ta interfaced samples present the
most relevant differences in their properties, while the TiN interfaced sample
can be considered as an intermediate situation. . . . . . . . . . . . . . 114
150
Bibliography
[1] G.E. Moore. Cramming more components onto integrated circuits. Elec-
tronics, 38(8):114–117, 1965.
[2] G. W. Burr, M. J. Breitwisch, M. Franceschini, D. Garetto, K. Gopalakr-
ishnan, B. Jackson, B. Kurdi, C. Lam, L. A. Lastras, A. Padilla, B. Rajen-
dran, S. Raoux, and R. S. Shenoy. Phase change memory technology. Jour-
nal of Vacuum Science & Technology B: Microelectronics and Nanometer
Structures, 28(2):223, 2010.
[3] S. Ovshinsky. Reversible electrical switching phenomena in disordered
structures. Physical Review Letters, 21(20):1450–1453, November 1968.
[4] S. Raoux and M. Wuttig. Phase change materials: science and applica-
tions. Springer Verlag, 2008.
[5] N. Yamada, E. Ohno, N. Akahira, K. Nishiuchi, K. Nagata, and M. Takao.
High speed overwritable phase change optical disk material. Japanese
Journal of Applied Physics, 26S4(Supplement 26-4):61–66, 1987.
[6] N. Yamada, E. Ohno, K. Nishiuchi, N. Akahira, and M. Takao. Rapid-
phase transitions of GeTe−Sb2Te3 pseudobinary amorphous thin films for
an optical disk memory. Journal of Applied Physics, 69(5):2849, 1991.
[7] M. Wuttig and N. Yamada. Phase-change materials for rewriteable data
storage. Nature Materials, 6(11):824–832, November 2007.
[8] D. Lencer, M. Salinga, and M. Wuttig. Design rules for phase-change
materials in data storage applications. Advanced Materials, 23(18):2030–
2058, May 2011.
151
BIBLIOGRAPHY BIBLIOGRAPHY
[9] G. Betti Beneventi. Characterization and modeling of phase-change mem-
ories. PhD Thesis, 2011.
[10] D. Ielmini and Y. Zhang. Analytical model for subthreshold conduction
and threshold switching in chalcogenide-based memory devices. Journal
of Applied Physics, 102(5):054517–054517–13, September 2007.
[11] D. Ielmini. Threshold switching mechanism by high-field energy gain
in the hopping transport of chalcogenide glasses. Physical Review B,
78(3):035308, 2008.
[12] J. A. Kalb. Crystallization kinetics in antimony and tellurium alloys used
for phase change recording. Doctoral thesis, RWTH Aachen University,
2006.
[13] J. W. Christian. The Theory of Transformations in Metals and Alloys.
Pergamon Press, 1975.
[14] K. F. Kelton, H. Ehrenreich, and D. Turnbull. Crystal nucleation in liquids
and glasses. In Solid State Physics, volume Volume 45, pages 75–177.
Academic Press, 1991.
[15] O. Penrose. The Becker-Dring equations for the kinetics of phase transi-
tions. August, 2001.
[16] D. Turnbull and J. C. Fisher. Rate of nucleation in condensed systems.
The Journal of Chemical Physics, 17(1):71, 1949.
[17] J. Willard Gibbs. The scientific papers of J. Willard Gibbs. Dover, New
York, 1961.
[18] D. T. Wu. Nucleation theory. Solid State Physics - Advances in Research
and Applications, Vol 50, 50:37–187, 1997.
[19] H. E. Kissinger. Reaction kinetics in differential thermal analysis. Ana-
lytical chemistry, 29(11):1702–1706, 1957.
152
BIBLIOGRAPHY BIBLIOGRAPHY
[20] S. Senkader and D. C. Wright. Models for phase-change of Ge2Sb2Te5
in optical and electrical memory devices. Journal of Applied Physics,
95(2):504, 2004.
[21] H.Y. Cheng, T.H. Hsu, S. Raoux, J.Y. Wu, P.Y. Du, M. Breitwisch,
Y. Zhu, E.K. Lai, E. Joseph, S. Mittal, R. Cheek, A. Schrott, S.C. Lai,
H.L. Lung, and C. Lam. A high performance phase change memory with
fast switching speed and high temperature retention by engineering the
GexSbyTez phase change material. In Electron Devices Meeting (IEDM),
2011 IEEE International, pages 3.4.1 –3.4.4, December 2011.
[22] A. Fantini, L. Perniola, M. Armand, J.F. Nodin, V. Sousa, A. Persico,
J. Cluzel, C. Jahan, S. Maitrejean, S. Lhostis, A. Roule, C. Dressler,
G. Reimbold, B. De Salvo, P. Mazoyer, D. Bensahel, and F. Boulanger.
Comparative assessment of GST and GeTe materials for application to
embedded phase-change memory devices. In Memory Workshop, 2009.
IMW ’09. IEEE International, pages 1–2, 2009.
[23] E. Gourvest, B. Pelissier, C. Valle, A. Roule, S. Lhostis, and S. Maitrejean.
Impact of oxidation on Ge2Sb2Te5 and GeTe phase-change properties.
Journal of The Electrochemical Society, 159(4):H373H377, 2012.
[24] L. Perniola, V. Sousa, A. Fantini, E. Arbaoui, A. Bastard, M. Armand,
A. Fargeix, C. Jahan, J.-F. Nodin, A. Persico, D. Blachier, A. Toffoli,
S. Loubriat, E. Gourvest, G. Betti Beneventi, H. Feldis, S. Maitrejean,
S. Lhostis, A. Roule, O. Cueto, G. Reimbold, L. Poupinet, T. Billon,
B. De Salvo, D. Bensahel, P. Mazoyer, R. Annunziata, P. Zuliani, and
F. Boulanger. Electrical behavior of phase-change memory cells based on
GeTe. IEEE Electron Device Letters, 31(5):488 –490, May 2010.
[25] T. Nonaka, G. Ohbayashi, Y. Toriumi, Y. Mori, and H. Hashimoto. Crystal
structure of GeTe and Ge2Sb2Te5 meta-stable phase. Thin Solid Films,
370(1):258–261, 2000.
153
BIBLIOGRAPHY BIBLIOGRAPHY
[26] N. Yamada and T. Matsunaga. Structure of laser-crystallized Ge2Sb2 +
xTe5 sputtered thin films for use in optical memory. Journal of Applied
Physics, 88(12):7020–7028, December 2000.
[27] Shinji Kohara, Kenichi Kato, Shigeru Kimura, Hitoshi Tanaka, Takeshi
Usuki, Kentaro Suzuya, Hiroshi Tanaka, Yutaka Moritomo, Toshiyuki
Matsunaga, Noboru Yamada, Yoshihito Tanaka, Hiroyoshi Suematsu, and
Masaki Takata. Structural basis for the fast phase change of ge[sub
2]Sb[sub 2]Te[sub 5]: Ring statistics analogy between the crystal and amor-
phous states. Applied Physics Letters, 89(20):201910, 2006.
[28] J. Akola and R. Jones. Structural phase transitions on the nanoscale:
The crucial pattern in the phase-change materials Ge2Sb2Te5 and GeTe.
Physical Review B, 76(23), December 2007.
[29] A. V. Kolobov, P. Fons, A. I. Frenkel, A. L. Ankudinov, J. Tominaga,
and T. Uruga. Understanding the phase-change mechanism of rewritable
optical media. Nature Materials, 3(10):703–708, September 2004.
[30] S. Hosokawa, T. Ozaki, K. Hayashi, N. Happo, M. Fujiwara, K. Horii,
P. Fons, A. V. Kolobov, and J. Tominaga. Existence of tetrahedral site
symmetry about Ge atoms in a single-crystal film of Ge2Sb2Te5 found
by x-ray fluorescence holography. Applied Physics Letters, 90(13):131913,
2007.
[31] K. S. Andrikopoulos, S. N. Yannopoulos, G. A. Voyiatzis, A. V. Kolobov,
M. Ribes, and J. Tominaga. Raman scattering study of the a-GeTe struc-
ture and possible mechanism for the amorphous to crystal transition. Jour-
nal of Physics: Condensed Matter, 18(3):965–979, January 2006.
[32] K.S. Andrikopoulos, S.N. Yannopoulos, A.V. Kolobov, P. Fons, and
J. Tominaga. Raman scattering study of GeTe and Ge2Sb2Te5 phase-
change materials. Journal of Physics and Chemistry of Solids, 68(5-
6):1074–1078, May 2007.
154
BIBLIOGRAPHY BIBLIOGRAPHY
[33] J. Akola and R. Jones. Binary alloys of Ge and Te: Order, voids, and the
eutectic composition. Physical Review Letters, 100(20), May 2008.
[34] R. Kojima, S. Okabayashi, T. Kashihara, K. Horai, T. Matsunaga,
E. Ohno, N. Yamada, and T. Ohta. Nitrogen doping effect on phase
change optical disks. Japanese Journal of Applied Physics, 37(Part 1, No.
4B):2098–2103, 1998.
[35] S. Y. Kim, S. J. Kim, H. Seo, and M. R. Kim. Complex refractive indices of
GeSbTe-Alloy thin films: Effect of nitrogen doping and wavelength depen-
dence. Japanese Journal of Applied Physics, 38(Part 1, No. 3B):1713–1714,
1999.
[36] S. Privitera, E. Rimini, and R. Zonca. Amorphous-to-crystal transition of
nitrogen- and oxygen-doped Ge2Sb2Te5 films studied by in situ resistance
measurements. Applied Physics Letters, 85(15):3044, 2004.
[37] K. Kim, J. Park, J. Chung, S. A. Song, M. Jung, Y. M. Lee, H-J Shin,
B. Kuh, Y. Ha, and J-S Noh. Observation of molecular nitrogen in n-doped
Ge2Sb2Te5. Applied Physics Letters, 89(24):243520–243520–3, December
2006.
[38] L. W. Fang, Z. Zhang, R. Zhao, J. Pan, M. Li, L. Shi, T. Chong, and
Y. Yeo. Fermi-level pinning and charge neutrality level in nitrogen-doped
Ge2Sb2Te5 : Characterization and application in phase change memory
devices. Journal of Applied Physics, 108(5):053708–053708–7, September
2010.
[39] H. Horii, J.H. Yi, J.H. Park, Y.H. Ha, I.G. Baek, S.O. Park, Y.N. Hwang,
S.H. Lee, Y.T. Kim, K.H. Lee, U.-In Chung, and J.T. Moon. A novel cell
technology using n-doped GeSbTe films for phase change RAM. In 2003
Symposium on VLSI Technology, 2003. Digest of Technical Papers, pages
177 – 178, June 2003.
[40] S.J. Ahn, Y.J. Song, C.W. Jeong, J.M. Shin, Y. Fai, Y.N. Hwang, S.H. Lee,
K.C. Ryoo, S.Y. Lee, J.H. Park, H. Horii, Y.H. Ha, J.H. Yi, B.J. Kuh,
155
BIBLIOGRAPHY BIBLIOGRAPHY
G.H. Koh, G.T. Jeong, H.S. Jeong, Kinam Kim, and B.I. Ryu. Highly
manufacturable high density phase change memory of 64Mb and beyond.
In Electron Devices Meeting, 2004. IEDM Technical Digest. IEEE Inter-
national, pages 907 – 910, December 2004.
[41] A. Fantini, V. Sousa, L. Perniola, E. Gourvest, J.C. Bastien, S. Maitrejean,
S. Braga, N. Pashkov, A. Bastard, B. Hyot, A. Roule, A. Persico, H. Feldis,
C. Jahan, J.F. Nodin, D. Blachier, A. Toffoli, G. Reimbold, F. Fillot,
F. Pierre, R. Annunziata, D. Benshael, P. Mazoyer, C. Vallee, T. Billon,
J. Hazart, B. De Salvo, and F. Boulanger. N-doped GeTe as performance
booster for embedded phase-change memories. In Electron Devices Meeting
(IEDM), 2010 IEEE International, pages 29.1.1 –29.1.4, December 2010.
[42] K. Kim, S. Choi, J. Chung, J. Lee, and S. Heo. Changes in chemical and
structural properties of phase-change material GeTe with nitrogen doping
and annealing. Japanese Journal of Applied Physics, 49(6):061801, June
2010.
[43] C. Peng, L. Wu, F. Rao, Z. Song, X. Zhou, M. Zhu, B. Liu, D. Yao,
S. Feng, P. Yang, and J. Chu. Nitrogen incorporated GeTe phase change
thin film for high-temperature data retention and low-power application.
Scripta Materialia, 65(4):327–330, 2011.
[44] G. Betti Beneventi, L. Perniola, V. Sousa, E. Gourvest, S. Maitrejean, J.C.
Bastien, A. Bastard, B. Hyot, A. Fargeix, C. Jahan, J.F. Nodin, A. Per-
sico, A. Fantini, D. Blachier, A. Toffoli, S. Loubriat, A. Roule, S. Lhostis,
H. Feldis, G. Reimbold, T. Billon, B. De Salvo, L. Larcher, P. Pavan,
D. Bensahel, P. Mazoyer, R. Annunziata, P. Zuliani, and F. Boulanger.
Carbon-doped GeTe: a promising material for phase-change memories.
Solid-State Electronics, 6566(0):197–204, November 2011.
[45] P. Noe, J. Y. Raty, G. E. Ghezzi, F. Hippert, E. Souchier, L. Perniola,
V. Sousa, A. Persico, E Henaff, M. Tessaire, C. Bichara, and S. Maitrejean.
Effect of dopant incorporation in phase change material. E\PCOS 2012
Proceedings, 2012.
156
BIBLIOGRAPHY BIBLIOGRAPHY
[46] G. E. Ghezzi, J. Y. Raty, S. Maitrejean, A. Roule, E. Elkaim, and F. Hip-
pert. Effect of carbon doping on the structure of amorphous GeTe phase
change material. Applied Physics Letters, 99(15):151906, 2011.
[47] D. A. Keen. A comparison of various commonly used correlation func-
tions for describing total scattering. Journal of Applied Crystallography,
34(2):172–177, April 2001.
[48] H. E. Fischer, A. C. Barnes, and P. S Salmon. Neutron and x-ray
diffraction studies of liquids and glasses. Reports on Progress in Physics,
69(1):233–299, January 2006.
[49] T. E. Faber and J. M. Ziman. A theory of the electrical properties of liquid
metals. Philosophical Magazine, 11(109):153–173, 1965.
[50] M. Micoulaut, J.-Y. Raty, C. Otjacques, and C. Bichara. Understanding
amorphous phase-change materials from the viewpoint of maxwell rigidity.
Physical Review B, 81(17), May 2010.
[51] J-Y. Raty, C. Otjacques, J-P. Gaspard, and C. Bichara. Amorphous struc-
ture and electronic properties of the ge1sb2te4 phase change material. Solid
State Sciences, 12(2):193–198, February 2010.
[52] J. Akola and R. Jones. Structure of amorphous Ge8Sb2Te11 GeTe-Sb2Te3
alloys and optical storage. Physical Review B, 79(13), April 2009.
[53] C. Bichara and J-Y. Raty. Temperature-induced density anomaly in te-
rich liquid germanium tellurides: p versus sp3 bonding. Physical Review
Letters, 95(26), December 2005.
[54] J. Akola, J. Larrucea, and R. Jones. Polymorphism in phase-change
materials: melt-quenched and as-deposited amorphous structures in
Ge2Sb2Te5 from density functional calculations. Physical Review B, 83(9),
March 2011.
157
BIBLIOGRAPHY BIBLIOGRAPHY
[55] T. Chattopadhyay, J. X. Boucherle, and H. G. VonSchnering. Neutron
diffraction study on the structural phase transition in GeTe. Journal of
Physics C: Solid State Physics, 20(10):1431–1440, April 1987.
[56] Alexander V. Kolobov, Paul Fons, Berangere Hyot, Bernard Andre, Junji
Tominaga, Yusuke Tamenori, Hideki Yoshikawa, and Keisuke Kobayashi.
Local structure of nitrogen in n-doped amorphous and crystalline GeTe.
Applied Physics Letters, 100(6):061910, 2012.
[57] K. B. Borisenko, Y. Chen, D. J.H. Cockayne, S. A. Song, and H. S. Jeong.
Understanding atomic structures of amorphous c-doped Ge2Sb2Te5 phase-
change memory materials. Acta Materialia, 59(11):4335–4342, June 2011.
[58] E. Cho, Y. Youn, and S. Han. Enhanced amorphous stability of carbon-
doped Ge2Sb2Te5 : Ab initio investigation. Applied Physics Letters,
99(18):183501–183501–3, November 2011.
[59] Eunae Cho, Seungwu Han, Dohyung Kim, Hideki Horii, and Ho-Seok
Nam. Ab initio study on influence of dopants on crystalline and amor-
phous Ge2Sb2Te5. Journal of Applied Physics, 109(4):043705–043705–10,
February 2011.
[60] Y.M. Lee, H.J. Shin, S.J. Choi, J.H. Oh, H.S. Jeong, K. Kim, and M.-
C. Jung. Nitrogen contribution to N-doped GeTe (N: 8.4at.%) in the
structural phase transition. Current Applied Physics, 11(3):710–713, May
2011.
[61] J-Y. Raty, P. Noe, G. E. Ghezzi, S. Maitrejean, C. Bichara, and F. Hip-
pert. Vibrational properties and stabilization mechanism of the doped gete
phase change material amorphous phase. Submitted to Phys. Rev. Lett.,
().
[62] X. Zhou, L. Wu, Z. Song, F. Rao, M. Zhu, C. Peng, D. Yao, S. Song,
B. Liu, and S. Feng. Carbon-doped Ge2Sb2Te5 phase change material:
A candidate for high-density phase change memory application. Applied
Physics Letters, 101(14):142104–142104–4, October 2012.
158
BIBLIOGRAPHY BIBLIOGRAPHY
[63] Xavier Biquard, Milos Krbal, Alexander V. Kolobov, Paul Fons, Robert E.
Simpson, Brangre Hyot, Bernard Andr, Junji Tominaga, and Tomoya
Uruga. Effect of doping on global and local order in crystalline GeTe.
Applied Physics Letters, 98(23):231907–231907–3, June 2011.
[64] B. Liu, Z. Song, T. Zhang, J. Xia, S. Feng, and B. Chen. Effect of n-
implantation on the structural and electrical characteristics of Ge2Sb2Te5
phase change film. Thin Solid Films, 478(12):49–55, 2005.
[65] G. E. Ghezzi, R. Morel, A. Brenac, N. Boudet, M. Audier, F. Fillot,
S. Maitrejean, and F. Hippert. Crystallization of Ge2Sb2Te5 nanometric
phase change material clusters made by gas-phase condensation. Applied
Physics Letters, 101(23):233113–233113–4, December 2012.
[66] N. Ohshima. Crystallization of germaniumantimonytellurium amorphous
thin film sandwiched between various dielectric protective films. Journal
of Applied Physics, 79(11):8357, 1996.
[67] X. Wei, S. Luping, Z. Chong, C. T.and Rong, and L.H. Koon. Thick-
ness dependent nano-crystallization in Ge2Sb2Te5 films and its effect on
devices. Japanese Journal of Applied Physics, 46(4B):2211–2214, April
2007.
[68] S. Privitera, S. Lombardo, C. Bongiorno, E. Rimini, and A. Pirovano.
Phase change mechanisms in Ge2Sb2Te5 . Journal of Applied Physics,
102(1):013516, 2007.
[69] J. Kalb, F. Spaepen, and M. Wuttig. Calorimetric measurements of phase
transformations in thin films of amorphous te alloys used for optical data
storage. Journal of Applied Physics, 93(5):2389, 2003.
[70] S. Raoux, Jean L. J.-S., and A. J. Kellock. Crystallization properties of
ultrathin phase change films. Journal of Applied Physics, 103(11):114310,
2008.
159
BIBLIOGRAPHY BIBLIOGRAPHY
[71] M. Zacharias and P. Streitenberger. Crystallization of amorphous super-
lattices in the limit of ultrathin films with oxide interfaces. Physical Review
B, 62(12):8391, 2000.
[72] R. E. Simpson, M. Krbal, P. Fons, A. V. Kolobov, J. Tominaga, T. Uruga,
and H. Tanida. Toward the ultimate limit of phase change in Ge2Sb2Te5 .
Nano letters, 10(2):414419, 2009.
[73] W. D. Nix. Mechanical properties of thin films. Metallurgical Transactions
A, 20(11):2217–2245, November 1989.
[74] H.-Y. Cheng, S. Raoux, C.-W. Cheng, and J. L. Jordan-Sweet. The crys-
tallization behavior of stressed Ge2Sb2Te5 phase-change material. MRS
Spring Meeting 2012, 2012.
[75] F. Oki, Y. Ogawa, and Y. Fujiki. Effect of deposited metals on the crys-
tallization temperature of amorphous germanium film. Japanese Journal
of Applied Physics, 8:1056, 1969.
[76] H. Homma, I. K. Schuller, W. Sevenhans, and Y. Bruynseraede. Inter-
facially initiated crystallization in amorphous germanium films. Applied
Physics Letters, 50(10):594, 1987.
[77] N. Ohshima. Structural analysis and crystallization studies of germani-
umantimony tellurium sputtered films on different underlayers. 83(10),
1998.
[78] M. H. Jang, S. J. Park, D. H. Lim, M.-H. Cho, Y. K. Kim, H.-J. Yi,
and H. S. Kim. Structural stability and phase-change characteristics of
Ge2Sb2Te5 nano-multilayered films. Electrochemical and Solid-State Let-
ters, 12(4):H151, 2009.
[79] S. Meister, H. Peng, K. McIlwrath, K. Jarausch, X. F. Zhang, and Y. Cui.
Synthesis and characterization of phase-change nanowires. Nano letters,
6(7):15141517, 2006.
160
BIBLIOGRAPHY BIBLIOGRAPHY
[80] D. Yu, J. Wu, Q. Gu, and H. Park. Germanium telluride nanowires and
nanohelices with memory-switching behavior. Journal of the American
Chemical Society, 128(25):8148–8149, 2006.
[81] S.-H. Lee, Y. Jung, and R. Agarwal. Highly scalable non-volatile and
ultra-low-power phase-change nanowire memory. Nature Nanotechnology,
2(10):626–630, September 2007.
[82] S.-H. Lee, Y. Jung, and R. Agarwal. Size-dependent surface-induced het-
erogeneous nucleation driven phase-change in Ge2Sb2Te5 nanowires. Nano
letters, 8(10):3303–3309, October 2008. PMID: 18778106.
[83] S. Raoux, C. T. Rettner, J. L. Jordan-Sweet, A. J. Kellock, T. Topuria,
P. M. Rice, and D. C. Miller. Direct observation of amorphous to crys-
talline phase transitions in nanoparticle arrays of phase change materials.
Journal of Applied Physics, 102(9):094305, 2007.
[84] Y. Zhang, S. Raoux, D. Krebs, L. E. Krupp, T. Topuria, M. A. Caldwell,
D. J. Milliron, A. Kellock, Philip M. Rice, J. L. Jordan-Sweet, and H.-S. P.
Wong. Phase change nanodots patterning using a self-assembled poly-
mer lithography and crystallization analysis. Journal of Applied Physics,
104(7):074312, 2008.
[85] H. S. Choi, K. S. Seol, K. Takeuchi, J. Fujita, and Y. Ohki. Synthesis of
size- and structure-controlled Ge2Sb2Te5 nanoparticles. Japanese Journal
of Applied Physics, 44(10), 2005.
[86] H. R. Yoon, W. Jo, E. H. Lee, J. H. Lee, M. Kim, K. Y. Lee, and Y. Khang.
Generation of phase-change Ge–Sb–Te nanoparticles by pulsed laser abla-
tion. Journal of non-crystalline solids, 351(43):3430–3434, 2005.
[87] G.-S. Park, J.-H. Kwon, M. Kim, H. R. Yoon, W. Jo, T. K. Kim, J.-M.
Zuo, and Y. Khang. Crystalline and amorphous structures of Ge–Sb–Te
nanoparticles. Journal of Applied Physics, 102(1):013524, 2007.
161
BIBLIOGRAPHY BIBLIOGRAPHY
[88] I. U. Arachchige, R. Soriano, C. D. Malliakas, S. A. Ivanov, and M. G.
Kanatzidis. Amorphous and crystalline GeTe nanocrystals. Advanced
Functional Materials, 21(14):2737–2743, July 2011.
[89] M. A. Caldwell, S. Raoux, R. Y. Wang, H.-S. Philip Wong, and D. J. Mil-
liron. Synthesis and size-dependent crystallization of colloidal germanium
telluride nanoparticles. Journal of Materials Chemistry, 20(7):1285, 2010.
[90] M. J. Polking, H. Zheng, R. Ramesh, and A. P. Alivisatos. Controlled
synthesis and size-dependent polarization domain structure of colloidal
germanium telluride nanocrystals. Journal of the American Chemical So-
ciety, 2011.
[91] R. Morel, A. Brenac, P. Bayle-Guillemaud, C. Portemont, and F. La Rizza.
Growth and properties of cobalt clusters made by sputtering gas-
aggregation. European Physical Journal D, 24(1-3):287–290, June 2003.
[92] H.R. Yoon, W. Jo, E. Cho, S. Yoon, and M. Kim. Microstructure and
optical properties of phase-change GeSbTe nanoparticles grown by pulsed-
laser ablation. Journal of Non-Crystalline Solids, 352(3637):3757–3761,
October 2006.
[93] W. K. Njoroge, H.-W. Woltgens, and M. Wuttig. Density changes upon
crystallization of Ge2Sb2.04Te4.74 films. Journal of Vacuum Science & Tech-
nology A: Vacuum, Surfaces, and Films, 20(1):230, 2002.
[94] T. P. Leervad Pedersen, J. Kalb, W. K. Njoroge, D. Wamwangi, M. Wuttig,
and F. Spaepen. Mechanical stresses upon crystallization in phase change
materials. Applied Physics Letters, 79(22):3597, 2001.
[95] D.-M. Smilgies. Scherrer grain-size analysis adapted to grazing-incidence
scattering with area detectors. Journal of Applied Crystallography,
42(6):1030–1034, October 2009.
[96] S. Meister, D. T. Schoen, M. A. Topinka, A. M. Minor, and Y. Cui. Void
formation induced electrical switching in phase-change nanowires. Nano
Letters, 8(12):4562–4567, December 2008.
162
BIBLIOGRAPHY BIBLIOGRAPHY
[97] V. Weidenhof, I. Friedrich, S. Ziegler, and M. Wuttig. Laser induced
crystallization of amorphous Ge2Sb2Te5 films. Journal of Applied Physics,
89(6):3168, 2001.
[98] L. H. Zhang, E. Johnson, and U. Dahmen. Observations of vacancy-
controlled decay of elastic strain caused by phase transformation of small
pb inclusions in al. Acta materialia, 53(13):36353642, 2005.
[99] K. J. Choi, S. M. Yoon, N. Y. Lee, S. Y. Lee, Y. S. Park, B. G. Yu, and
S. O. Ryu. The effect of antimony-doping on Ge2Sb2Te5 , a phase change
material. Thin Solid Films, 516(23):88108812, 2008.
[100] M. Wuttig, D. Lsebrink, D. Wamwangi, W. Wenic, M. Gillessen, and
R. Dronskowski. The role of vacancies and local distortions in the design
of new phase-change materials. Nature Materials, 6(2):122–128, December
2006.
[101] J.A. Kalb, M. Wuttig, and F. Spaepen. Calorimetric measurements of
structural relaxation and glass transition temperatures in sputtered films
of amorphous te alloys used for phase change recording. Journal of Mate-
rials Research, 22(03):748–754, March 2011.
[102] A. Marmier, K. Kohary, and D. C. Wright. Determination of the
anisotropic elastic properties of Ge1Sb2Te4. Applied Physics Letters,
98:231911, 2011.
[103] S. Raoux, B. Munoz, H.-Y. Cheng, and J. L. Jordan-Sweet. Phase tran-
sitions in Ge–Te phase change materials studied by time-resolved x-ray
diffraction. Applied Physics Letters, 95(14):143118, 2009.
[104] M. Chen, K. A. Rubin, and R. W. Barton. Compound materials for
reversible, phasechange optical data storage. Applied Physics Letters,
49(9):502–504, September 1986.
[105] C. V. Thompson and R. Carel. Stress and grain growth in thin films.
Journal of the Mechanics and Physics of Solids, 44(5):657673, 1996.
163
BIBLIOGRAPHY BIBLIOGRAPHY
[106] B.J. Mrstik, H.L. Hughes, P. Gouker, R.K. Lawrence, and P.J. McMarr.
The role of nanoclusters in reducing hole trapping in ion implanted oxides.
IEEE Transactions on Nuclear Science, 50(6):1947 – 1953, December 2003.
[107] A.P. Hammersley. FIT2D: an introduction and overview. ESRF Internal
Report, ESRF97HA02T, 1997.
[108] X. Qiu, J. W. Thompson, and S. J. L. Billinge. PDFgetX2: a GUI-
driven program to obtain the pair distribution function from x-ray powder
diffraction data. Journal of Applied Crystallography, 37(4):678–678, July
2004.
164
Acknowledgments
First of all, I would like to thank with all my heart my Thesis Director
Francoise Hippert. With her exceptional energy and dedication to work, not to
mention her sense of humor, she helped and encouraged me constantly through
these three years like no-one else could do. I learnt much from her, and we had
such wonderful (even when hard!) times between synchrotrons, office work at
night and conferences that filled those three years with beautiful memories I
will never forget. My deep gratitude goes also to the co-Director of this thesis,
Sylvain Maitrejean, that was able to take care of me even if time was never
enough, always with an optimistic aptitude that can make any problem easier.
I would also like to thank with all my heart Frederic Fillot, who taught me so
much and with whom I had many wonderful conversations that I really treasure.
I wish we had more time to spend together.
A great thank goes to all the PCM team of CEA Leti. First of all to the project
leader Veronique Sousa and to Pierre Noe, not only for their competence in work,
patience, dynamism and optimism, but also for being the best companions that
one can desire when night never comes in Finland. I thank Luca Perniola, great
worker and solid reference for all the italians PhD students, for the fruitful
discussions and for the good laughs. Thanks to Olga Cueto for helping me
with learning COMSOL, to Alain Persico, Carine Jahan, Jean-Francois Nodin,
Christope Vallee, Philippe Michallon, and to everyone that I could have forgotten
to add. Thanks to Anne Roule, Mathieu Petit and Ewen Henaff for the thin films
samples depositions. I deeply thank Robert Morel and Ariel Brenac for their
165
Acknowledgments
wonderful clusters deposition and for the good time we had while characterizing
them, as well as for their enormous knowledge and competence that helped me so
much. I am particularly grateful to Jean-Yves Raty, who can find a theoretical
explanation for all our experimental evidences even before they are found (really
impressive!). Thanks to Jean-Paul Simon, Nathalie Boudet and Jean-Francois
Berard, that were our precious contact on the BM02 D2NT beamline at the
ESRF, and to Erik Elkaim that supported us so much at SOLEIL.
My gratitude goes to all the PhD students of the laboratory, old and new. I
would like to thank in particular Audrey who is one of the kindest persons I
have ever met, and my wonderful cobureau Kavita, Sylvia, Raul and Bilel with
whom I divided for almost two years a small office room that felt like home.
I deeply thank all the PhD students of the PCM group, each of whom placed
its brick in building the PCM project: Giovanni e Stefania, Emmanuel, Jean
Claude, Eddie, Gabriele, Quentin, Manan, the sweet Sarra and everyone else.
I hope I did not forget anyone, and I apologize if I did so. Thank you all, I wish
you all the best.
I would like to make some more personal acknowledgments and the best
thing to do is to write them in the own language of the persons I would like to
thank.
Il primo ringraziamento va alla mia splendida, sgangherata famiglia greno-
blese, passata e presente: Lia, Ramo, Chiara, Gan, Vera, Paolo, GBB e Ste-
fania, Giova1, Simeon, Eric, Caro, Giova2, Ricky, Carlo e Clio+Panga, Cus,
Fil... Senza di voi sarei disperata (e probabilmente sotto un ponte (ˆ.-)). Siete
davvero la mia famiglia, e questa tesi esiste grazie a voi.
Je remercie encore une fois Francoise, pas dans son role de Directrice de these
mais plutot comme une chere amie qui a partage avec mois le travail de ces trois
annees. Merci pour tout, je n’ai pas les mots pour t’exprimer ma gratitude.
Merci aussi a Robert, l’extraordinaire mari de Francoise, pour sa gentillesse, son
esprit agreable, les conversations en italien et l’habilite infaillible de trouver les
meilleurs restos a San Francisco et a Helsinki. Merci a Sylvain qui a ete tou-
jours pret a m’aider quand j’en ai eu besoin, et qui ma toujours donne des bons
conseils. Merci a Fred, qui m’a aide beaucoup pendent mes premiers jours au
166
Acknowledgments
Leti. Merci encore a l’equipe PCM, formee des personnes tellement agreables
que travailler avec eux c’est un plaisir. Un merci et un biz a Pilou et a Geor-
gette! Je remerci beaucoup Robert et Ariel pour la gentillesse et les moments
de bonheur, et le tres sympa Jean-Yves sans lequel cette these n’aurait que trois
chapitres.
Merci encore a mes amies Audrey, Kavita, Sylvia, aux thesards du labo et du
projet, et merci aussi aux thesards DCOS ”collegues de Lia” pour les repas bien
passes. Merci a Radekko, Cornelia, Pablo et a tout les autres ”etrangers de
Grenoble” qui forment une communaute tres chaleureuse.
Ringrazio gli amici lasciati a Milano, che non mi hanno dimenticata in questi
tre anni: la Piera, che resta la mia insostituibile migliore amica, Deiv e i suoi
cappelli che amo immensamente, Albe, Eleele e Qc (e il terzo incomodo), Vera,
Noja, Rezzo, il Niggah, Vanish, Dade, Tumji e Sunday, la Brini, Matt e il sem-
pregiovane Frank. Un pensiero pieno di gratitudine va a Marco, che mi ha
accompagnata alla soglia di questo viaggio, e alla sua famiglia. Ringrazio con
tantissimo affetto le meravigliose amiche del giardino, Gloria, Rubina, Elisa,
Stefania e Tosca, che tra alti e bassi e nonostante la lontananza restano un
caposaldo fondamentale della mia vita. Grazie di cuore a tutta la famiglia Ar-
brun/Maurino, ed in particolare a Matilde, Ezio e Floriana, per avermi accolta
con tanto affetto.
Rigrazio con tutto il cuore il mio Stefano, compagno della difficile stesura di
questa tesi e compagno della mia vita, senza il quale nulla avrebbe senso. Gra-
zie per essere stato con me in questa avventura, e per tutte quelle che ancora ci
aspettano.
Il ringraziamento piu sentito e profondo va alla mia piccola ma solida famiglia:
Ernesto, Antonella, Nike, Serena e Gio, Angelo e Silvia, i miei nonni. In par-
ticolare, tutta la mia gratitudine va a mamma e papa. Il vostro amore, la
comprensione e il supporto che mi avete sempre dato sono la base su cui poggia
la mia vita.
167
Abstract
AbstractPhase Change Memories (PCM) are one of the best candidates for the next generation of non volatile
memories. A great research effort is still needed in order to optimize the properties of phase change (PC)
materials which are used in PCM devices. In particular, doping has been demonstrated to improve retention
in devices. Moreover, a study of the effect of scaling and interface material on PC materials properties is still
an open research field. In this context, the first part of the thesis is dedicated to investigate the local structure
of C or N doped amorphous GeTe. The impact of doping is observed experimentally with the appearance of a
new peak in the pair distribution function of doped GeTe, indicating the formation of a bond at a new distance
that is absent in the undoped amorphous material. The presence of new environments involving carbon and
nitrogen is confirmed through ab initio simulations. The subject of the second part of this thesis is the impact
of confinement on Ge2Sb2Te5 (GST) crystallization mechanism. Nano-sized clusters of GST have been made
by sputtering, deposited and then studied through X-ray diffraction using synchrotron radiation. The crys-
talline clusters experience a tensile strain that can be ascribed to the effect of the embedding Al2O3 matrix.
Their crystallization temperature has been found to be only 25C higher than the one observed for a thin
film of GST of 10 nm deposited under the same conditions. This result is positive for the future development
PCM because it indicates that the scaling effect on the crystallization temperature in phase change material
can be small. The third and last part of the thesis is dedicated to the investigation of the interface material
effect on the crystallization temperature of GeTe and GST thin films through reflectivity and X-ray diffraction
measurements. In both GeTe and GST film 100 nm thick interfaced with Ta the crystallization temperature
is higher than in the case of TiN or SiO2 interface. Such an interface effect on relatively thick films was never
reported before. The results suggest that the SiO2 /GeTe interface is energetically favorable for the nucleation
and growth of grains with a preferred orientation and that nucleation and growth mechanisms are different for
different interface materials.
ResumeLes memoires a changement de phase sont l’un des candidats les plus prometteurs pour la prochaine
generation de memoires non-volatiles. Un intense effort de recherche est requis pour optimiser les materiaux
a changement de phase (PC) utilises dans ces memoires. En particulier, il a ete demontre que le dopage
ameliore les proprietes de retention des dispositifs. Par ailleurs, l’etude des effets de reduction de taille et des
effets des materiaux d’interface sur les proprietes des materiaux a changement de phase est encore un sujet de
recherche ouvert. Dans ce contexte, la premiere partie de la these est dediee a l’investigation de la structure
locale de GeTe amorphe dope avec C ou N. L’effet du dopage sur la structure a ete observe experimentalement
via l’apparition d’un nouveau pic dans la fonction de distribution de paires de GeTe dope, ce qui montre
la formation d’une nouvelle liaison interatomique absente dans le materiau non dope. La presence de nou-
velles configurations incluant le carbone et l’azote a ete confirmee par des simulations ab initio. L’objet de la
deuxieme partie de la these est l’influence de la reduction de taille sur la cristallisation de Ge2Sb2Te5 (GST).
Des agregats nanometriques de GST ont ete fabriques par pulverisation puis deposes et etudies par diffrac-
tion des rayons X en utilisant le rayonnement synchrotron. Dans l’etat cristallise une tres forte deformation
positive des agregats est observee et attribuee a la matrice d’Al2O3 qui entoure les agregats. La temperature
de cristallisation des agregats est de 25C plus elevee que celle d’un film de GST de 10 nm depose dans les
memes conditions. Ce resultat est encourageant pour les futurs developpements des memoires a changement
de phase car il montre que l’effet de reduction de taille sur la temperature de cristallisation peut-etre faible.
La troisieme et derniere partie de la these est dediee a l’investigation des effets des materiaux d’interface sur la
temperature de cristallisation de films minces de GeTe et GST par des mesures de reflectivite et de diffraction
des rayons X. Pour les deux materiaux, la temperature de cristallisation de films de 100 nm est plus grande
pour une interface avec du Ta que pour une interface avec du TiN ou du SiO2 . Une difference aussi marquee
n’etait jamais montre auparavant. Les resultats suggerent que l’interface SiO2 /GeTe est energetiquement
favorable pour la nucleation et la croissance de grains avec une orientation preferentielle et que les mecanismes
de nucleation et croissance sont differents pour differents materiaux d’interface.
169